Wahid Digital

Category: AI Search Optimization

  • Google SERP Poisoning: How It Is Disrupting Rankings and AI Overview Sources

    Google SERP Poisoning: How It Is Disrupting Rankings and AI Overview Sources

    Google’s search results have become increasingly difficult to interpret for SEO professionals. Since September 2026, several SEO data providers have reported an unusual pattern: Google appears to be returning significantly different search results to automated data collectors than those seen in normal search experiences.

    The phenomenon is increasingly being referred to as SERP poisoning.

    What makes this particularly interesting is that the disruption does not appear to be limited to traditional organic search rankings. Research published by Nozzle suggests that AI Overview sources, People Also Ask results and related search suggestions are also experiencing substantial changes.

    In fact, one of the most significant findings is that the retention of previously cited AI Overview sources has dropped from 68.6% to just 35.8% in poisoned SERPs.

    This raises an important question for the SEO and Generative Engine Optimisation (GEO) industry: are we measuring genuine changes in Google’s search ecosystem, or are some of our tracking tools receiving a distorted version of it?

    Let’s look at the findings.

    What Is Google SERP Poisoning?

    SERP poisoning describes a phenomenon where automated search data collectors receive complete Google search results pages containing results that are substantially different from the usual results for a particular query.

    Unlike a conventional scraping failure, these pages can look completely normal.

    They load successfully, contain organic results, and may include other familiar SERP features. However, many of the websites that normally rank for the keyword are suddenly missing.

    Instead, unrelated or loosely relevant websites appear in their place.

    This creates a problem for rank-tracking platforms because their systems can record these results as genuine ranking movements.

    For example, a website ranking in position three could suddenly disappear from the top 10 in a tracking report, even though its actual visibility in Google’s normal search experience has not changed.

    The issue becomes particularly concerning when these inaccurate observations are stored and used for historical ranking comparisons.

    A Real-World Example: Asbestos Removal Glasgow

    I recently came across an interesting example while investigating the query “asbestos removal Glasgow”.

    I compared two screenshots for the same search query.

    The first screenshot was captured through a SERP data collection tool. The second came from an actual Google search performed manually.

    The differences were striking.

    1. The rank-tracking SERP returned completely unrelated results

    The SERP captured by the tracking tool contained organic listings related to:

    • Carbon removal and carbon removal credits.
    • EU carbon removal standards.
    • Nitrogen dioxide (NO₂) removal policies.
    • Woodland removal and forestry policies.
    • Carbon removal technology and markets.

    These topics have very little relevance to someone searching for an asbestos removal contractor in Glasgow.

    Even the related search suggestions appeared to follow this unrelated carbon and environmental removal theme.

    2. The actual Google search returned relevant results

    When I searched for the same query manually, Google returned a completely different experience.

    The organic listings included relevant asbestos removal companies, while the Local Pack displayed Glasgow-based businesses offering asbestos-related services.

    The related search suggestions were also much more closely aligned with the original transactional intent.

    3. Local and paid results remained comparatively stable

    Another interesting observation was that the Local Pack and sponsored listings remained relevant in the rank-tracking screenshot.

    This is consistent with a wider pattern identified in Nozzle’s research, where different SERP components experienced very different levels of disruption.

    The important distinction: This example does not independently establish why Google returned different results. However, it provides a practical illustration of the kind of discrepancy that SEO professionals should investigate when rank-tracking data looks abnormal.

    Nozzle’s Research: Analysing 42.5 Million Google SERPs

    One of the most comprehensive investigations into this phenomenon comes from Nozzle.

    Its research, published by Derek Perkins and updated on 3 November 2026, analyses Google’s search results across multiple countries, devices and geographic targeting levels.

    The dataset includes:

    • 42.5 million SERPs collected since 13 September 2026.
    • 58 countries.
    • Desktop, iPhone and Android devices.
    • National, regional and more granular location targeting.

    Nozzle established a historical baseline using search results collected between 20 August and 9 September 2026.

    It classified a SERP as poisoned when fewer than half of a keyword’s usual websites appeared anywhere within the results page being evaluated.

    How widespread is the issue?

    The findings are significant.

    MetricNozzle’s finding
    SERPs analysed since 13 September42.5 million
    Average poisoning across page 1 or page 227.6%
    Previous baseline0.28%
    Peak poisoning level68.6%
    Date of peak2 November 2026
    Page 2 poisoning compared with page 11.5×

    The 68.6% peak relates specifically to SERPs where both pages were evaluated. It is not the percentage of all Google searches affected.

    Looking at page one separately, Nozzle reported a 15.0% poisoning rate across the study period, with a daily peak of 34.8% on 27 October.

    Another interesting finding is that page two was poisoned approximately 1.5 times as often as page one.

    In 56% of cases where page two was poisoned, page one still contained its usual results.

    This means a rank-tracking report focused exclusively on page one might miss a substantial part of the disruption.

    The Most Interesting Finding: AI Overview Sources Are Also Affected

    This is where Nozzle’s research becomes particularly relevant to the GEO and AI search industry.

    While most discussions around SERP poisoning have focused on traditional organic rankings, Nozzle also analysed how other Google SERP components behaved.

    One of the most striking findings concerns AI Overview sources.

    According to the research, the retention of previously appearing AI Overview sources declined from 68.6% in the historical baseline to 35.8% in poisoned SERPs.

    <figure>
    <figcaption><strong>AI Overview source retention: baseline vs poisoned SERPs</strong></figcaption>

    MeasurementSource retention
    Historical baseline68.6%
    Poisoned SERPs35.8%
    Absolute decline32.8 percentage points
    Relative decline47.8%

    </figure>

    In other words, previously observed AI Overview sources were retained considerably less often in the poisoned SERPs.

    This represents a relative decline of approximately 47.8%.

    However, there is an important distinction here: this is a measurement of source retention, not a measurement of citation accuracy.

    It does not mean that 47.8% of AI Overview citations are incorrect, nor does it prove that Google is producing lower-quality AI-generated answers.

    It tells us that the sources appearing in AI Overviews are considerably less consistent with those previously observed for the same keywords.

    An even more fascinating observation: AI Overviews appeared more frequently

    There is another detail in Nozzle’s data that deserves attention.

    The presence of AI Overviews actually increased in the poisoned SERP dataset.

    AI Overview metricBaselinePoisoned SERPs
    SERPs displaying AI Overview sources49.9%59.6%
    Previous sources retained68.6%35.8%

    This creates an interesting contrast.

    AI Overviews were present on a larger share of poisoned SERPs, but the sources they cited were considerably less consistent with their previous appearances.

    It suggests that the impact is not simply that AI Overviews disappear from search results.

    Instead, the continuity of their source selection is disrupted.

    For anyone monitoring AI Overview visibility, this is a particularly important distinction.

    Tracking whether an AI Overview appears is not necessarily the same as tracking whether the same websites continue to receive citations.

    The Impact Goes Beyond Organic Search

    Nozzle also compared the retention of other SERP components.

    The results show that different parts of Google Search experienced very different levels of disruption.

    Retention of previously observed SERP elements

    SERP componentBaselinePoisoned SERPs
    Organic results88.2%30.0%
    People Also Ask87.4%26.1%
    People Also Search For87.9%39.8%
    AI Overview sources68.6%35.8%
    Local Pack86.0%72.4%
    Video79.8%67.4%
    Image Pack84.7%72.7%
    Knowledge Panel87.3%78.8%
    Discussions and Forums90.1%82.0%
    Shopping93.2%93.8%
    Paid Ads69.6%71.8%

    Source: Nozzle research, data through 3 November 2026.

    A few observations stand out.

    Organic results experienced substantial disruption. Only 30% of previously observed organic websites were retained in the poisoned SERPs, compared with 88.2% in the baseline.

    People Also Ask was also heavily affected. Its retention declined from 87.4% to 26.1%.

    Local Pack results were comparatively more stable. Although retention declined from 86.0% to 72.4%, this was a considerably smaller change than the one observed in organic results.

    Paid ads and shopping results remained particularly stable. Their retention rates actually increased slightly.

    This suggests that SERP poisoning does not affect every component of Google Search equally.

    It is worth noting that these figures measure how much of the previously observed content was retained, rather than the accuracy or relevance of every individual result.

    Which Websites Are Appearing More Frequently in Poisoned SERPs?

    Nozzle also investigated which domains appeared more frequently in poisoned search results.

    The findings indicate a greater presence of social platforms and user-generated content websites.

    DomainOther SERPsPoisoned SERPs
    Facebook19.3%27.4%
    YouTube10.6%14.2%
    Instagram6.1%8.9%
    Quora1.8%3.9%
    MapQuest1.7%3.0%
    Etsy0.6%1.6%
    Reddit17.0%15.7%

    Facebook experienced the largest increase, appearing in 27.4% of poisoned SERPs compared with 19.3% of other SERPs for the same tracked keywords.

    YouTube, Instagram and Quora also became more prominent.

    Interestingly, Reddit moved in the opposite direction, declining slightly from 17.0% to 15.7%.

    This suggests that poisoned SERPs are not simply replacing established organic websites with every type of user-generated content. The distribution varies by platform.

    Is SERP Poisoning Affecting Every Country Equally?

    Nozzle’s country-level analysis also reveals significant geographical variation.

    Among the countries included in its published breakdown:

    CountryPoisoned SERPs
    South Africa38.0%
    Austria36.2%
    United Arab Emirates32.0%
    New Zealand30.4%
    Switzerland29.5%
    Singapore27.6%
    Germany16.3%
    India22.8%
    United Kingdom10.3%

    These are percentages from Nozzle’s own tracked sample, not estimates of every Google search performed in those countries.

    India’s sample contained 370,581 judged SERPs, while the UK sample contained approximately 4.13 million.

    The difference is worth investigating further, especially for SEO professionals monitoring international rankings.

    Why Might Google Be Returning These Results?

    An important question is why this is happening.

    Nozzle’s interpretation is that Google is returning misleading results to suspected automated data collectors as a response to large-scale AI scraping and the collection of search results for AI systems.

    This is a plausible explanation for the pattern, but it has not been publicly confirmed by Google.

    In an October 2026 update, Semrush also acknowledged the issue as an industry-wide problem affecting automated search data providers.

    Semrush noted that some automated requests were receiving SERPs that appeared normal but differed from what typical users would see. It also stated that Google had not publicly commented on the issue.

    The underlying mechanism and Google’s intentions therefore remain unconfirmed.

    It is important to distinguish between the observed behaviour, the industry’s interpretation of it and the explanation for why it happens.

    What Does This Mean for SEO and GEO Professionals?

    The implications go beyond temporary ranking fluctuations.

    For SEO professionals, the immediate concern is the reliability of third-party ranking data.

    A sudden drop in keyword rankings does not necessarily mean a website has lost its organic visibility.

    For GEO professionals, the implications may be even more interesting.

    If the sources appearing in AI Overviews are also affected in SERP data collected by automated tools, then AI visibility measurements could potentially reflect a distorted representation of Google’s source-selection behaviour.

    This makes it important to distinguish between:

    • AI Overview presence: Does an AI Overview appear for the query?
    • Source retention: Are previously cited websites still appearing?
    • Source diversity: Which new websites are being cited?
    • Citation visibility: Is a particular website still being referenced?

    These metrics should not be treated as interchangeable.

    How Should SEOs Validate Their Ranking Data?

    Based on these findings, I would recommend a few practical steps before reacting to unexpected ranking movements.

    1. Validate unexpected ranking drops manually.

    Check the actual Google search results for the affected keyword, using the same target country, language and device wherever possible.

    2. Compare third-party data with Google Search Console.

    If a website suddenly disappears from a rank-tracking report but continues receiving impressions and clicks for the affected queries, investigate the discrepancy before assuming that rankings have genuinely collapsed.

    Search Console is not an exact replacement for rank tracking, but its performance data provides a useful independent signal.

    3. Monitor page two as well as page one.

    Nozzle’s research shows that page two is affected more frequently. Monitoring only the first page may therefore leave some anomalies undetected.

    4. Compare multiple data providers.

    A disagreement between tracking platforms can be a useful warning signal. However, multiple providers may also be exposed to similar collection issues, so agreement alone does not guarantee accuracy.

    5. Separate AI Overview presence from source retention.

    For GEO reporting, monitor whether the AI Overview appears and which domains it cites. A change in one does not necessarily imply a change in the other.

    6. Avoid making strategic decisions based on isolated ranking movements.

    Look for persistent changes across multiple observations before modifying content, technical SEO or link-building strategies.

    Is the SERP Poisoning Issue Still Ongoing?

    The situation appears to be evolving.

    In its 6 November 2026 update, Ahrefs reported that its page-one data had largely recovered and stabilised.

    However, it was still filtering abnormal results for some keywords on page two and beyond. Ahrefs also reported that the proportion of affected results varied between data providers, with its earlier estimates ranging from 5% to 20%.

    Semrush has separately described the issue as industry-wide and outlined changes to its data collection and validation processes.

    This reinforces an important point: the problem may evolve over time, and different data providers may identify, filter and report abnormal results differently.

    The Nozzle figures discussed in this article represent its research snapshot through 3 November 2026, not a permanent or universal measurement of Google’s search results.

    Final Thoughts: Are We Measuring Google’s Real Search Experience?

    SERP poisoning presents an interesting challenge for the SEO industry.

    For years, ranking reports have been a fundamental part of measuring organic search performance. As AI Overviews and other AI-powered search features become increasingly important, the industry is also investing more heavily in monitoring citations and source visibility.

    However, these measurements depend on the reliability of the underlying search data.

    Nozzle’s research highlights that a search results page can look perfectly normal while containing a substantially different set of results.

    More importantly, its findings suggest that AI Overview source retention is also affected, with a 47.8% relative decline compared with the historical baseline.

    I believe this deserves further investigation from both SEO and GEO researchers.

    The next question is whether this disruption can also be observed consistently across different SERP APIs, manual Google searches and AI Overview tracking platforms.

    Are we measuring genuine changes in Google’s search results, or are some of our tools measuring an altered version of the SERP?

    That distinction could become increasingly important as the industry moves towards measuring visibility across both traditional search and AI-generated answers.

    Sources and Further Reading

    1. Nozzle – Poisoned Google SERPs (updated 3 November 2026)
    2. Ahrefs – An Update on Google’s Poisoning of Search Results (updated 6 November 2026)
    3. Semrush – Poisoned SERPs: An Update on Google Ranking Data
    4. Search Engine Journal – Analysis: Google Is Poisoning Search Results for Rank Trackers
    5. Search Engine Journal – How Rank Trackers Say They Handle Suspect Google Results

    Research note: Statistics in this article are attributed to the respective providers. Nozzle’s findings describe its own tracked SERP sample and should not be interpreted as a census of all Google searches.

  • What happens to CTR when an AI Overview shows up

    What happens to CTR when an AI Overview shows up

    AI Overviews do not affect organic clicks uniformly. An analysis of non-branded search queries for an anonymous service brand revealed a clear difference between queries where an AI Overview appeared and those where it was already present.

    When an AI Overview appeared, impressions remained broadly stable, but clicks declined and click-through rate fell noticeably.

    Queries that already had an AI Overview showed a different pattern. Their impressions and clicks declined together, while CTR remained broadly unchanged.

    The findings suggest that the arrival of an AI Overview may have a more immediate relationship with organic click-through rate than its continued presence. Here is how the analysis was conducted, what the results suggest, and where the limitations matter.

    The setup

    We compared search performance across two snapshots taken at different points in the year. The brand and underlying keyword dataset remain anonymous, and the findings are presented in qualitative terms rather than as exact performance figures.

    The analysis began with non-branded keywords identified by a SERP rank-tracking platform as displaying an AI Overview in the later snapshot. These keywords were matched with available Google Search Console data to examine changes in impressions, clicks and click-through rate.

    The matched keywords were divided into two groups:

    • Newly appearing AI Overview: the query did not display an AI Overview in the earlier snapshot but did in the later one.
    • AI Overview present throughout: the query displayed an AI Overview in both snapshots.

    The second group provides a useful comparison, although it is not a true control group. The analysis did not include a separate group of queries that remained consistently free of AI Overviews.

    To reduce the influence of other factors, the comparison focused on keywords with relatively stable tracked rankings. A seasonal segment with an unusually sharp decline in demand was also excluded from the primary comparison.

    These safeguards helped make the groups more comparable, although they cannot eliminate every source of bias.

    The method: separating demand from click-through rate

    A decline in organic clicks can have different explanations. Search demand may have fallen, impressions may have changed, or a smaller proportion of searchers may have clicked through to the website.

    To distinguish these effects, we decomposed the change in clicks into an impression-driven component and a CTR-driven component.

    [
    \text{Impression effect} =
    (\text{Later impressions} – \text{Earlier impressions})
    \times \text{Earlier CTR}
    ]

    [
    \text{CTR effect} =
    \text{Later impressions}
    \times
    (\text{Later CTR} – \text{Earlier CTR})
    ]

    This approach changes one factor at a time, allowing the two components to reconcile with the overall click change.

    The distinction matters because declining impressions and declining CTR represent different challenges. The former may reflect changes in demand or visibility, while the latter indicates that impressions are generating fewer organic clicks.

    However, even when CTR declines, the calculation does not establish why users are less likely to click.

    Finding: when an AI Overview appears, CTR comes under pressure

    Queries that gained an AI Overview maintained broadly stable impression levels, yet organic clicks declined noticeably.

    The difference was primarily associated with weaker click-through rate rather than a corresponding reduction in impressions.

    This suggests that the website continued to receive search visibility, but a smaller share of searchers clicked through to its organic listing.

    One possible explanation is that an AI Overview satisfies some users’ information needs before they visit an external website. Another is that the additional SERP feature changes the prominence and position of traditional organic results.

    Both explanations are plausible, but the available observations cannot isolate the contribution of each factor.

    The important takeaway is that stable impressions do not necessarily translate into stable organic traffic when the search results page changes.

    Finding: when the AI Overview is already present, CTR is more stable

    The comparison group followed a different pattern.

    For queries where an AI Overview was present in both snapshots, impressions and clicks moved in broadly similar directions, while CTR remained relatively stable.

    This suggests that the decline in organic clicks was more closely associated with reduced search activity than with a further deterioration in click-through rate.

    It also highlights why measuring AI Overview presence alone may be insufficient. The point at which a feature first appears could be associated with a different performance pattern from its continued presence.

    That distinction deserves further investigation across additional datasets and search categories.

    [Optional chart: qualitative comparison of newly appearing and previously established AI Overviews. Show directional trends only, without exact values.]

    Finding: group averages can tell the wrong story

    Average CTR is influenced by more than changes in individual keywords. It can also change when the composition of a group shifts.

    For example, a group may contain queries that usually attract relatively high CTR alongside queries that attract very few clicks. If impressions shift disproportionately towards either type, the group’s average can change even when individual keyword performance has not improved or deteriorated in the same way.

    The analysis highlighted several important cases.

    Seasonality can exaggerate the apparent effect.

    A seasonal segment experienced a substantial decline in search activity after its peak period. Because these queries typically attracted a relatively high click-through rate, their changing contribution affected the group’s overall average.

    Removing the segment produced a more balanced comparison and reduced the risk of attributing seasonal behaviour to AI Overviews.

    CTR can improve even as traffic falls.

    Some seasonal queries recorded lower impressions and clicks alongside stronger CTR. One possible explanation is a shift in search intent: casual researchers may disappear after the peak period, leaving users with a more immediate need for the service.

    This remains a hypothesis rather than a demonstrated cause. Changes in the keyword mix may also account for part of the movement.

    An apparently better average can conceal weaker keyword performance.

    A specialist segment showed an improvement in average CTR that was largely explained by changes in the distribution of impressions.

    Queries with negligible click activity lost visibility, changing the weight of different keywords in the group average. At individual keyword level, performance was frequently less positive than the aggregate result suggested.

    The broader lesson is to investigate changes within comparable keyword groups before interpreting an overall average as evidence of improvement or decline.

    Finding: high-volume queries matter most

    The aggregate CTR decline was not distributed evenly across all keywords.

    A relatively small group of high-volume queries contributed disproportionately to the overall change. Generic searches and location-modified service queries were particularly important in explaining the pattern.

    Long-tail queries with little search activity had less influence on the aggregate result.

    This has practical implications for monitoring. A website-wide CTR average may conceal a meaningful change in the queries that contribute most to organic traffic.

    The analysis also highlighted a potential difference between tracked rankings and the positions reported in Search Console. Tracked rankings remained relatively stable by design, while impression-weighted Search Console positions weakened for the group where AI Overviews newly appeared.

    This is consistent with the possibility that changes to the results-page layout affect organic visibility, even when conventional rank tracking suggests limited movement.

    However, the observations cannot distinguish the effect of the AI Overview itself from the effect of the changing layout around it.

    Keyword-level analyses also did not establish a consistent independent effect across the dataset. The aggregate pattern should therefore be interpreted as an association that warrants further investigation, not as a universal outcome for every query.

    What SEO teams should do

    Monitor changes in AI Overview presence. Record when a feature first appears, rather than tracking presence alone. This creates a clearer basis for investigating changes in organic performance.

    Separate impression and CTR movements. Determine whether a traffic decline is associated with weaker search demand, reduced visibility or a lower propensity to click. Each points towards a different potential response.

    Control for rankings and seasonality. Compare queries with broadly stable rankings and investigate seasonal segments separately before attributing traffic changes to a SERP feature.

    Analyse comparable groups. Review performance within keyword segments instead of relying exclusively on averages. Changes in the mix of queries can conceal the underlying behaviour.

    Prioritise high-volume queries. Investigate important generic and location-based searches when AI Overviews appear. Review organic visibility, available citation information and Search Console performance together.

    Extend the measurement framework. Where possible, record whether a brand is cited within the AI Overview. Citation presence could help distinguish changes in organic click behaviour across different types of AI-generated results.

    Limitations

    This analysis covers an anonymous brand’s non-branded search performance across a limited number of snapshots. It should not be treated as representative of all industries, websites or search intents.

    Seasonality and changes in user demand may have influenced the observed movements. The comparison group provides useful context but does not establish what would have happened without AI Overviews.

    Rank tracking records observed search results under specific tracking conditions, whereas Google Search Console aggregates performance across users, devices and locations. The datasets therefore do not necessarily represent identical search experiences.

    The analysis also cannot establish whether AI Overview citations influence clicks, as citation status was not incorporated into the comparison.

    Further research across additional websites, search categories and observation periods would help determine how consistently these patterns occur.

    The central conclusion is that AI Overview presence alone is not enough to explain organic performance. Understanding when the feature appears, how impressions and CTR change, and how individual query groups behave provides a more useful foundation for SEO decision-making.

  • ChatGPT Atlas User-Agent: Why You’re Suddenly Seeing It in Your Server Logs

    ChatGPT Atlas User-Agent: Why You’re Suddenly Seeing It in Your Server Logs

    Website owners and SEO specialists scanning server logs in late 2025 are noticing an unusual new entry:

    ChatGPT%20Atlas/2025xxxx CFNetwork/xxxx.x Darwin/xx.x.x

    This is not a bot. It’s the user-agent of ChatGPT Atlas, OpenAI’s new AI-powered browser now rolling out globally for Mac OS Users on paid plans.

    Note: As per experimentation with my own website and data from my server logs, the above user agent is only accessing my Favicon and Logo.

    When using Atlas browser, the rest of the content and web-page is accessed through the below user agent which appears to be a chrome instance. This would make it hard to track and block ChatGPT Atlas browser. (Source: chrisrcook.com)

    Mozilla/5.0 (Macintosh; Intel Mac OS X 10_12_3) AppleWebKit/123.45 (KHTML, like Gecko) Chrome/142.0.0.0 Safari/537.36

    While the AI industry is discussing Atlas for its “agent mode” and automation features, very few developers or SEOs understand what this user-agent means when it shows up in access logs. This article explains it in a clear, technical but concise way as much as possible.

    What Exactly Is ChatGPT Atlas?

    ChatGPT Atlas is a standalone browser built by OpenAI that merges web browsing with live AI assistance. Instead of switching between a browser and ChatGPT, users browse the web inside ChatGPT.

    As more users adopt Atlas, your server logs will begin showing requests coming from this new browser.

    What the User-Agent Actually Means

    When you see a string like this:

    ChatGPT%20Atlas/20251112345678123 CFNetwork/1234.123.123 Darwin/12.3.4

    Here’s the technical breakdown:

    • ChatGPT%20Atlas → A normal space-encoded identifier (%20 = space)
      The request came from the ChatGPT Atlas browser, not Chrome, Safari, or Firefox.
    • Build Number (20251112345678123) → This is a unique version/build identifier.
      It represents the internal build OpenAI is running.
    • CFNetwork/xxxx → Apple’s networking framework used by macOS apps.
      This confirms the user is browsing through the macOS version of ChatGPT Atlas.
    • Darwin/xx.x.x → Darwin kernel version, the Unix foundation of macOS.

    Note: It’s not clear yet what’s the role of this user agent as I was unable to find any information on this user agent on the internet except the below one.

    If you want to know more about this user agent, go to useragents.io

    Important for SEOs:

    ✔ It is not a crawler.

    A Unique Feature: How Atlas Displays Search Results

    One of the biggest UX differences between using ChatGPT in Atlas and in traditional browsers: the way search results are displayed.

    1. AI-Generated Answer With Direct Citations (Top Section)

    When a user asks something (e.g., “best project management tool for 2025”), Atlas instantly generates a structured answer.
    At the top, it shows:

    • Key points
    • Summaries
    • A neatly formatted table
    • Clickable source cards from the web-pages it used (This is not shown if you use ChatGPT in other browsers such as Chrome or Safari). However, this source card will only show up up ChatGPT uses web based retrieval for the prompt. If the web-based retrieval is not used then clickable source card with the below mentioned featured will not show up.

    These source cards appear above the AI answer.

    2. “Globe Icon” Tab = Classic Search Results (Like Google)

    Users can tap the 🌐 Globe icon to switch to raw web search results. This feature is only available when you are using ChatGPT through Atlas browser. You will not see this feature when you use chatGPT through Google Chrome or any other browser.

    This tab shows:

    • Ten blue-link results
    • Titles
    • Meta descriptions
    • URLs
    • Exactly like a Google SERP

    So Atlas delivers two layers of search:

    3. Image, Video and News Tabs – Additional Search Layers

    Next to the Globe icon, Atlas also provides Image, Video and News tabs, similar to Google’s vertical search options.

    These tabs allow users to switch between content formats without leaving the chat:

    • Image Tab: Infographics, dashboards, screenshots, comparison graphics
    • Video Tab: Reviews, product demos, tutorials, walkthroughs mainly from Youtube
    • News Tab: Latest announcements, industry updates, product news

    Together, these layers make Atlas feel like a complete search engine, not just an AI assistant.

    Why This Matters for SEO Professionals

    1. Atlas Is Becoming a New “Referrer Type”

    Expect to see more sessions from the below two user agents in your server logs:

    ChatGPT%20Atlas/2025xxxx CFNetwork/xxxx.x Darwin/xx.x.x
    Mozilla/5.0 (Macintosh; Intel Mac OS X 10_12_3) AppleWebKit/123.45 (KHTML, like Gecko) Chrome/142.0.0.0 Safari/537.36

    Its not clear yet what’s the role of the first one. However, the second one signifies real users viewing your site inside ChatGPT Atlas browser.

    2. AI-Driven Summaries Equal New Competition for SERP Visibility

    Atlas directly summarises your content similar to chatGPT but the responses may vary.
    Structured content, Q&A formatting and clear headings increase the chance of being selected as a cited source.

    3. Content Must Be AI-Readable and Human-Readable

    Atlas does both:

    • Summarises your page
    • Shows raw search results
    • Displays visual and news layers

    Your optimisation strategy must therefore serve all layers:

    • AI summaries
    • Web results
    • Images
    • Videos
    • News

    Clean content structuring, internal linking and topical depth are now essential.

    Source: https://openai.com/index/introducing-chatgpt-atlas/

    Thanks for reading this Article!

  • KPIs for the Future of AI Search Optimization: The Metrics That Matter in 2025 and beyond

    KPIs for the Future of AI Search Optimization: The Metrics That Matter in 2025 and beyond

    The KPIs for SEO in the AI Search era require a dramatic shift in mindset and measurement. Instead of traditional metrics like clicks and keyword rankings, future-forward KPIs focus on visibility, influence, and impact within AI-driven environments where answers, not just links, rule the SERP landscape.

    Why the Old KPIs Are Fading

    Classic SEO KPIs such as click-through rates and organic sessions are declining in value as search evolves toward more answer-centric experiences powered by generative AI. In these environments, users may receive complete answers and brand impressions without ever visiting your website. That means “zero click” performance and AI driven citations become essential signals of success.

    The New Core KPIs for AI SEO Success

    Here are the most critical modern KPIs for AI-driven search optimization in 2025, reflecting current industry consensus and the latest strategic frameworks:

    Legacy KPI Modern AI Native KPI Explanation
    Organic Traffic Answer Visibility / Zero-Click Presence Measures whether your brand/content appears in AI Overviews, ChatGPT, Perplexity, featured snippets even without generating clicks.
    Keyword Rankings Embedding Relevance Score Tracks semantic match between your content and user queries via vector embeddings essential for LLM visibility.
    Click-Through Rate (CTR) Brand Recall & AI Citation Count Number and quality of times your brand/content is mentioned in AI answers across platforms. Build authority and awareness beyond blue links.
    Bounce Rate / Session Duration Engagement Quality & Interaction Metrics This moves beyond flawed metrics like bounce rate, which often mislabels a successful single-page visit as a failure. Instead, it tracks positive user actions that signal value, such as scroll depth (>75%), time-on-page (active reading), content shares, saves, and direct interactions with on-page tools like calculators or chatbots.
    Backlinks Vector Index Presence Rate Measures how much of your content is indexed in vector databases, a new form of technical SEO to maximize retrievability.
    Crawl Budget Optimization Crawl Efficiency & Bot Interaction This evolves the traditional practice of optimizing for Googlebot. By analyzing server logs, you now track how all bots—including new AI crawlers like Google-Extended and others AI bots interact with your site. The goal is to ensure these bots can efficiently access your most important content to use as a source for AI generated answers, while not wasting resources on low value pages.
    Conversions Multi-Touch Contribution & Attribution Rate Tracks how many conversions your content influences across multiple touchpoints, including AI responses, snippets, and journey assists.

    Essential Emerging Metrics Explained

    • Chunk Retrieval Frequency: Monitors how often modular segments (“chunks”) of your content are fetched by AI in response to user prompts. A higher frequency means better visibility in retrieval-augmented search environments.​
    • Embedding Relevance Score: Assesses the similarity between your content embedding and query embedding, signifying semantic value and answer accuracy.​
    • AI Citation Count: Tallies your brand’s total references within AI-generated answers, crucial for trust and authority as LLMs increasingly curate the SERP.​
    • Vector Index Presence Rate: Quantifies the portion of your site indexed into AI vector databases miss it, and you miss retrieval opportunities.​
    • Retrieval Confidence Score: Reflects AI system certainty in selecting your content. High confidence boosts inclusion in machine-generated results.​

    Influence and Impact: The Next Layer

    In the AI search pipeline, it’s no longer enough just to rank, you must optimize for each stage:

    • Intent: Is your content recognized as relevant to user needs by AI engines? Embedding quality, recall, and vector indexing matter.​
    • Influence: How often is your brand mentioned, cited, or surfaced in machine-led answers? Track citation, attribution, and share of visible answer spaces.​
    • Impact: Are users converting, engaging deeply, or searching for your brand after exposure? Multi-touch attribution and engagement metrics are vital.​

    Strategic Recommendations for AI SEO

    • Structure content for embedding clarity and semantic relevance.​
    • Invest in visibility tools for tracking your performance across AI surfaces, snippets, and answer engines.​
    • Monitor not just what gets clicked, but what gets shown, cited, and summarized.​
    • Update reporting dashboards to include both traditional and AI-native KPIs for holistic measurement.​

    The future of AI SEO is all about being seen, cited, and influential. Start tracking these next-generation KPIs now to future-proof your strategy in a landscape where answers, not just clicks, define success.​

    Sources:

  • Why AI Search crawlers can’t access client side rendered content and how to fix it

    With the rise of AI-driven search engines such as ChatGPT, Perplexity, and Gemini, understanding how your website content is rendered has become more important than ever.

    Traditional SEO best practices already emphasised indexability for Googlebot but AI crawlers behave differently.

    Recent tests, including a case study on aisearchoptimization.in
    , reveal that AI crawlers often fail to process content loaded using Client-Side Rendering (CSR). In other words, if your data appears only after clicking a button or through JavaScript, AI bots might never see it.

    1. How AI Crawlers Work

    AI crawlers such as those from ChatGPT, Perplexity, and Gemini typically:

    • Fetch a page’s HTML source code directly from the server.
    • Parse visible text and meta data.
    • Use large language models (LLMs) to summarise, categorise, and store information for future retrieval.

    However, these crawlers generally don’t execute JavaScript.
    That means if your content is loaded dynamically after page load, it doesn’t exist in the HTML snapshot the crawler receives.

    2. The Problem with Client-Side Rendering

    Client-Side Rendering (CSR) is when content is generated on the user’s browser via JavaScript frameworks such as React, Vue, or Angular.

    <div id="ai-stats"></div>
<script>
  fetch('/api/ai-data')
    .then(response => response.json())
    .then(data => document.getElementById('ai-stats').innerHTML = data.stats);
</script>

    To a user, this looks fine the stats appear after a button click.
    But to an AI crawler (or even a basic SEO bot), the HTML it fetches might look like this:

    <div id="ai-stats"></div>

    Result? Empty content.

    This is exactly what happens with some AI crawlers, the crawlers can’t see the dynamically loaded numbers unless they’re present in the HTML at load time.

    3. Why It Matters for AI Search Optimisation

    AI search engines rely on content visibility rather than traditional link crawling.
    If your critical data (like pricing, reviews, or research stats) is missing from the HTML:

    • AI tools won’t cite your content in responses.
    • Your brand’s authority in generative answers decreases.
    • Organic AI traffic (from ChatGPT or Perplexity) may drop dramatically.

    In short, if AI can’t read it, you don’t exist in the AI index.

    4. The Solution: Server-Side Rendering (SSR) or Static HTML

    To make your content AI-friendly:

    Approach Description Benefit
    Server-Side Rendering (SSR) Render full content on the server before sending it to the browser. AI crawlers can read all content instantly.
    Static HTML Export Pre-render your React/Vue components as HTML (e.g., with Next.js getStaticProps). No JS execution required by crawlers.
    Hybrid Rendering Combine SSR for content and CSR for interactive UI elements. Balance SEO visibility and user experience.

    5. Quick Fix Checklist

    ✅ Ensure all essential text appears in the raw HTML source (right-click → “View Page Source”).
    ✅ Use Server-Side Rendering (SSR) for your main content.
    ✅ Avoid loading key text only after button clicks or fetch calls.
    ✅ Test pages with AI crawlers.
    ✅ Validate visibility using AI-Search engines.

    6. Key Takeaway

    To be visible in the age of AI Search, your content must live in the HTML, not in JavaScript.
    AI crawlers are getting smarter, but they still can’t “click”, “scroll”, or “wait for scripts” like human users can.

    The simplest way to future-proof your site for AI visibility is to adopt Server-Side Rendering (SSR) or static HTML generation.

    🧠 If AI can’t read your content, it can’t recommend you.

    Make Your Brand Visible to AI Search Engines

    Don’t let client-side rendering hide your valuable content. Ensure AI crawlers can read and recommend your website.

    Contact Us

  • Robots.txt and noindex robots meta tag might not be the best way to block AI Search Crawlers

    I recently conducted an experiment on my test website. I applied the noindex meta robots tag to every page of the site. Then, I prompted several AI search engines to extract specific information from the website and observed the responses.

    For this scenario, let’s consider my test website to be domain.in.

    The prompt I used was: “What services does https://www.domain.in/ offer as a business? Please check the website directly before answering.”

    Key Findings

    • ChatGPT: Accessed the site’s content, as indicated by the ChatGPT-User/1.0 agent in my server logs (+https://openai.com/bot), and accurately quoted the requested information.
    • Perplexity: Did not retrieve any content from the website, suggesting it honors the noindex meta tag. Its response explicitly stated that direct website access is unavailable and that no content from the test site appears in search results.
    • Claude AI: Successfully obtained the required answers. The server logs showed the user agent Claude-User/1.0; +Claude-User@anthropic.com.
    • Google AI Mode: Generated fabricated information unrelated to the actual site content. This indicates it primarily relies on Google’s search index during its query process.
    • Deepseek AI: Could not access any content from the test website. Its response specified that it doesn’t browse the web directly and depends entirely on search results.
    • Qwen AI: Managed to retrieve content from the website, but its server logs revealed it used a HeadlessChrome browser rather than a specific user agent.

    This shows to me that adding noindex meta robots tag is not a safe way to block web-pages from AI Search engines.

    So after doing the above I had another test website where i added the below in my robots.txt which basically disallowed all the bots from accessing the website.

    User-agent: *
    Disallow: /

    However, even after adding the script, I used the below prompt to ask question about this new test website.

    Prompt: what services does https://www.domain.com/ as a business provide. Please check the website now and then only let me know. as they have updated their website What makes them different and why should i choose them.

    ChatGPT (specifically ChatGPT-User Agent as appearing in the Server Logs) was able to access my website and was able to quote the information as i asked in the prompt.

    I also tested the same prompt in other AI Answer engines but all the rest were not able to fetch any content from the website. Please see the findings below for more info.

    Claude AI: It specifically said that its unable to access the website directly as it appears to be blocked by robots.txt rules. This shows that it abides by the robots.txt rules.

    Perplexity AI: It wasnt able to get any information from the website. It implies that it respect the robots.txt rule.

    Deepseek: It wasnt able to get any information too. It gave the following response with some related information.

    Based on the search results provided, I do not have specific information about the services offered by https://www.domain.in/, as this particular website was not included in the search results. However, I can provide a general overview of what a typical AI Search Optimization Agency might offer based on the industry trends and common services described in the search results.

    It just shows that Deepseek relies on Search results from certain search engines.

    Google AI Mode: Wasn’t able to provide any information from the website.

    Qwen AI: Wasn’t able to provide any information from the website.

    This implies to me that even when you use robots.txt Disallow rule, its upto the AI Search Engine bot to respect it or not. So if you’re serious about this, you should use services like Cloudflare to block these bots which actively block these bots by identifying their IP Address and other criteria.

  • How to access website server logs in Hostinger Hpanel

    Do you know that you can access your website server logs or access logs in Hostinger Hpanel even for Starter Plans?

    If not then please have a read and at the end you would be able to access the Server Logs and able to find any 4XX and 5XX errors from your logs.

    Please follow the below steps as mentioned.

    1. First Login to your Hostinger account.
    2. Select your Hosting Package and Click on Manage
    1. Now select the website from the dropdown which appears in top portion of left hand vertical menu.
    2. Click on Analytics which appears on the second vertical menu from the left as shown below.
    Hostinger Hpanel Left hand menu
    1. Now click on Access logs Tab as shown
    Fig: Access Logs Tab inside the rectangular box

    This is were you will get see all your logs. You will be able to see the time when your website was accessed, IP Address, Country, Device/User Agent and Response Time.

    You can access what kind of bot accessed your website. For AI search engine optimization, we can specifically look at user agents such as GPTBot, ChatGPT-User, Google-Extended, ClaudeBot, PerplexityBot and many others. You can get an idea which pages these bots are requesting and which they don’t. This will give you an idea about the pages served in AI search results on these platform. Additionally you can also check the response time column to get an idea if the response time is healthy or certain pages are taking long time to respond to these bot requests.

    Note: You can’t export the Server logs from here and can only access a maximum of Last 7 days log data.

    Finding Error Logs

    1. Go through the Error Code 5xx tab to find all the pages having internal server errors. This will help you to find and resolve the issue which these AI bots are facing while accessing your web-pages
    2. At last, Go through the Error Code 4xx tab to find any broken pages. This will help you to either fix these pages or redirect these broken links to closely related web-pages.

    Once you have access to server logs, you can run various analysis on the data and find various issue and resolve it. If you are interested on how to do the analysis, please go through this Log Files Analysis guide by Matt Diggity.