Key takeaways
- The audit applies 58 surface-specific checks to three products: 19 to ChatGPT Search, 21 to Google AI Overviews and 18 to Google AI Mode.
- Each of the three also carries the same 30 shared measurement records, giving 49, 51 and 48 records respectively.
- ChatGPT’s checks revolve around three separate crawlers or roles, real bot evidence and how citations attach to sentences.
- Google’s checks revolve around index status, snippet eligibility, Search Console data and, for AI Mode, multi-turn follow-ups.
- Visibility, citation and real visits are always measured as three different things.
Three surfaces, three mechanisms
ChatGPT Search, Google AI Overviews and Google AI Mode all produce answers with links, so they are easy to lump together. They work differently, they expose different evidence and they are controlled by different settings on your side. Treating them as a single “AI search” channel hides exactly the differences that explain why you appear in one and not in another.
| Topic | ChatGPT Search | Google AI Overviews | Google AI Mode |
|---|---|---|---|
| Records in the catalog | 49 (30 shared, 19 specific) | 51 (30 shared, 21 specific) | 48 (30 shared, 18 specific) |
| Where it lives | Inside the ChatGPT conversation | On top of a Google results page | A separate conversational mode in Google |
| Main crawler or data path | OAI-SearchBot and web search tooling | Googlebot and Google Search index | Googlebot and Google Search index |
| Appears for every query? | Only when the model searches | No; it triggers for some searches | Only inside AI Mode sessions |
| Owner data source | Server logs, GA4 referrals | Search Console | Search Console (shared with AI Overviews) |
| Distinctive risk | Confusing bots and roles | Mistaking a missing trigger for a missing citation | Judging a conversation from its first answer only |
ChatGPT Search: 19 specific checks
ChatGPT is both a chat product and a search product. Whether a particular answer used the web at all is a fact that has to be recorded, not assumed. When it does search, several parties and settings determine whether your page is read and cited.
Three roles, three preferences
OpenAI documents separate crawlers for separate roles. OAI-SearchBot is used for search, GPTBot is used for model training, and ChatGPT-User fetches a page when a person asks for it. The audit reports each as its own item: CHATGPT-17 for the search crawler’s robots permission, CHATGPT-20 for the user-initiated role and CHATGPT-21 for the training preference. The training preference is shown as information and never as part of the score, because allowing or refusing training is a legitimate policy decision that has nothing to do with search visibility.
Permission is not the same as access
CHATGPT-17 reads your robots file. CHATGPT-18 and CHATGPT-19 ask a different question: did the bot really reach your page, and if not, which layer stopped it? A robots rule, a firewall and an error response are three different layers, and each needs a different fix. The checks match requests against OpenAI’s official bot identity and your server records instead of trusting a user-agent string.
Search, tools and modes
Several records keep apart things that are easy to confuse. A search tool that was requested is not the same as a search that completed (CHATGPT-31). The search that was permitted is not necessarily the search that ran (CHATGPT-28). Normal search, Deep Research and other modes are different series (CHATGPT-34), and the consumer product is different from the model interface (CHATGPT-27). Each distinction exists because blending them leads to conclusions about a product nobody actually measured.
Citations and what counts as one
CHATGPT-22 and CHATGPT-32 look at how links appear in an answer. A link can be a sentence citation, an entry in a source list or just navigation. The audit records the role of each, and checks that citation markers attach to the correct text fragment and the correct source. A source listed beneath an answer that no sentence relies on is not counted as a citation.
From citation to visit
CHATGPT-26 and CHATGPT-35 use your own analytics. Appearing in an answer and receiving a visit are different events. Where you share Google Analytics 4 data, visits and conversions that arrive from ChatGPT are reported on their own, so that nobody has to infer traffic from citation counts or the reverse.
Conditions that cannot be known
Some conditions are not disclosed. CHATGPT-24 stores the personalization settings that are known and flags measurements whose condition is unknown. CHATGPT-29 records which country or city information was used and how certain it is. CHATGPT-30 separates the influence of connected apps or uploaded files from open web results. Being open about what is unknown is part of the method.
ChatGPT Search: surface-specific checks (CHATGPT-17 to CHATGPT-35)
| Record | Check | What it examines |
|---|---|---|
CHATGPT-17 | OAI-SearchBot robots permission | Records whether your robots file allows ChatGPT's search crawler and which rule the decision rests on. |
CHATGPT-18 | Real bot IP and log evidence | Matches the bot's requests against OpenAI's official bot identity and your server logs to confirm it really fetched the target page. |
CHATGPT-19 | WAF and access layers | Distinguishes whether the crawler is blocked by the firewall, by a robots rule or by the server response, so the right layer gets fixed. |
CHATGPT-20 | ChatGPT-User role | Reports the ChatGPT-User bot, which fetches a page at a person's request, as information separate from search and training bots. |
CHATGPT-21 | GPTBot training preference | Shows your preference for GPTBot using content for model training as information separate from search visibility, and never as part of the score. |
CHATGPT-22 | Sources, sentence attribution and navigation | Records whether each link in an answer appears as a sentence citation, a source-list entry or a navigation link. |
CHATGPT-23 | Use of web search | Records whether ChatGPT actually searched the web while composing an answer, separately for consumer and model interfaces. |
CHATGPT-24 | Chat personalization | Stores the personalization settings that are known for a session and marks measurements whose condition is unknown. |
CHATGPT-25 | Content HTML access | Checks whether page content can be read by the ChatGPT bot, using the evidence from our own browser comparison. |
CHATGPT-26 | GA4 referrals and conversions | Uses the owner's Google Analytics data to separate visits and conversions that arrive from ChatGPT from citation counts. |
CHATGPT-27 | Model-interface research | Keeps results from the ChatGPT model interface apart from consumer ChatGPT results as a distinct surface and mode. |
CHATGPT-28 | Search selection and source scope | Records whether the search that was permitted and the search that actually ran are logged as separate facts. |
CHATGPT-29 | Approximate and precise location evidence | Records which country or city information was used to obtain a result and labels how certain each location is. |
CHATGPT-30 | Connected apps and uploaded files | Separates the influence of connected apps or uploaded files on an answer from what came from open web results. |
CHATGPT-31 | Tool request versus real execution | Reports a requested search tool and a search event that truly completed as two different things. |
CHATGPT-32 | Citation markers matching text | Checks whether citation markers in an answer attach to the correct text fragment and to the correct source. |
CHATGPT-33 | Query, filter and source list | Keeps the query used, the permission filter, the tool and the source list in a traceable record. |
CHATGPT-34 | Search, Deep Research and other modes | Makes sure the results of search, Deep Research and other modes are never mixed together. |
CHATGPT-35 | Separating real visits from citations | Treats visibility, citation and actual visits to your site as separate measures rather than interchangeable ones. |
Google AI Overviews: 21 specific checks
AI Overviews appear above or within Google search results for some queries. Because they sit on top of Google Search, the foundation is Google’s index and its snippet rules. The first thing the audit establishes is whether an overview appeared at all (AIO-21). A page cannot be cited in a feature that did not trigger, so a missing citation and a missing feature are different findings.
Eligibility starts with ordinary search
According to Google’s published guidance, a page needs to be indexed and eligible to show a snippet to be used as a supporting link in its AI features, and there is no extra technical requirement beyond that. That makes the classic checks decisive: Googlebot access through both robots rules and your CDN (AIO-17), the index status and the canonical Google selected (AIO-18), and preview directives such as nosnippet or max-snippet (AIO-19). AIO-20 then reports the effective preference your site has set for generative AI use in Google Search, with its scope and the chain of settings that produced it.
Organic rank is not an AI source
AIO-24 compares a page’s organic position with its appearance as an AI source without assuming that one predicts the other. In practice, some pages that rank well are never cited in an overview, and some pages that rank lower are. The audit records both facts and avoids drawing a causal line between them.
Search Console data, handled carefully
Google Search Console is the owner’s source of truth, and the catalog has more records about it than about any other external dataset: AIO-25 to AIO-28 and AIO-33 to AIO-35. They exist because the data is easy to misread. Totals can differ between a table and a chart depending on how the data is aggregated and which address is final. Zero in an export may be a true zero or hidden missing data. Country, device and page breakdowns overlap and cannot be added up. Recent data may be preliminary. A report that ignores these traps produces impressive-looking numbers that do not survive scrutiny.
Reading the answer itself
AIO-22 and AIO-32 record what the answer shows: the links in the expanded panel and the difference between the first card and the expanded answer. AIO-30 reports how often overviews appeared across a set of searches and, among the answers that formed, how often your site was cited. Two rates, with two different denominators, tell a far more useful story than one blended percentage.
Content, structure and special sources
AIO-23 verifies that important content is available in the mobile rendering. AIO-29 checks that structured data matches the visible content. AIO-31 records which input types are actually supported, and AIO-36 keeps standard Google results apart from experimental Labs results. AIO-37 applies only to sites that offer products or local businesses and checks that their profiles and source information are represented.
Google AI Overviews: surface-specific checks (AIO-17 to AIO-37)
| Record | Check | What it examines |
|---|---|---|
AIO-17 | Googlebot and CDN access | Compares robots permission with real Googlebot access, including through the content delivery network, as two separate questions. |
AIO-18 | Index status and selected canonical | Records whether the page is in Google's index and which canonical address Google selected, using the owner's dated data. |
AIO-19 | Snippet eligibility | Reads the preview directives that allow or restrict the page from being shown as a citation in Google AI Overviews. |
AIO-20 | Generative AI preference in Search | Reports the effective preference your site has set for generative AI use in Google Search, with its scope and the chain that produced it. |
AIO-21 | Whether an AI Overview appeared | Checks, for a successfully captured Google search, whether an AI Overviews section was actually generated. |
AIO-22 | Expanded link panel | Records the visible links in the expanded source panel of AI Overviews together with the state of the panel. |
AIO-23 | Rendered content on mobile | Verifies that key content is reachable in the page's mobile view as it is rendered in a browser. |
AIO-24 | Organic rank and AI source | Compares a page's rank in organic results with whether it appears as a source in AI Overviews, without assuming one predicts the other. |
AIO-25 | Search Console generative AI impressions | Imports the owner's Search Console generative AI data with the report scope and filters attached. |
AIO-26 | Country, device and page breakdown | Carries Search Console data by country, device and page, and avoids adding up overlapping totals. |
AIO-27 | Search type and preliminary data | Separates text from multimodal searches in Search Console and labels data that is not yet final. |
AIO-28 | Why data is absent in Search Console | Separates possible reasons for missing data in the generative AI report from reasons that are verified. |
AIO-29 | Structured data and visible content | Checks whether structured markup on the page matches what the user can actually see. |
AIO-30 | Trigger and visibility rate | Reports how often AI Overviews appeared across searches and, in the answers that formed, how often your site was cited. |
AIO-31 | Text, image and multimodal queries | Records which input formats Google AI Overviews truly supports and labels any format that is not supported. |
AIO-32 | Starting card and expanded answer | Compares the text and sources in the first card of an AI Overview with those in the expanded answer. |
AIO-33 | Preference chain and URL scope in Search Console | Checks whether AI preferences set in Search Console really cover the pages they are meant to cover. |
AIO-34 | Search Console totals and final URL | Reconciles differences between the table total and the chart total by using the correct aggregation method and the final address. |
AIO-35 | Missing-value loss in exports | Examines whether zeros in a Search Console file are true zeros or hidden missing data. |
AIO-36 | Standard search versus experimental Labs | Keeps Google's standard search results apart from experimental Labs results. |
AIO-37 | Product and local business sources | For sites that only offer products or local businesses, checks that profile and source information is represented in Google's results. |
Google AI Mode: 18 specific checks
AI Mode is a conversational mode in Google. The user asks a question and then keeps asking. That multi-turn quality is what makes its measurement different. A single question and a single answer is a poor description of how people use it, so the audit measures a conversation as a sequence of turns.
Turns, not just answers
AIMODE-17 confirms the result came from a separate AI Mode session. AIMODE-18 records the opening question and every follow-up as separate turns with their order and history. AIMODE-19 compares which brands and sources enter or leave the answer after each follow-up. AIMODE-26 requires the follow-up branches to be planned in advance and stored with their history, and AIMODE-27 makes sure each turn’s sources are tied only to that turn’s answer and passage. Without these rules, a brand that appears in turn three could wrongly be credited to turn one.
Links, panels and coverage
AIMODE-20 separates inline links within an answer from sources that are only listed in the panel. AIMODE-21 documents which sub-topics an answer covers and which it leaves out. This is often the most actionable finding in AI Mode: if answers consistently omit a sub-topic that your site covers well, you know which page needs to become clearer.
Honest limits on what is visible
AI Mode may run background sub-queries that the interface does not show. AIMODE-22 records only those the interface openly shows, and AIMODE-31 states an information limit for the rest. AIMODE-34 stops data shared between AI Overviews and AI Mode in Search Console from being presented as a standalone AI Mode metric, because that would claim precision Google does not provide. AIMODE-32 matches AI Overviews and AI Mode results under equal conditions and reports their difference as a separate rate.
Inputs, modes and cards
AIMODE-28 records whether the input was voice, an image or text. AIMODE-29 keeps experimental or special task modes in separate report series. AIMODE-30 separates the source named on a shopping or local card from the attribution in plain text. AIMODE-23, AIMODE-24, AIMODE-25 and AIMODE-33 cover search eligibility, the generative AI control scope, the starting condition of a session and the chain of property-level settings.
Google AI Mode: surface-specific checks (AIMODE-17 to AIMODE-34)
| Record | Check | What it examines |
|---|---|---|
AIMODE-17 | Whether AI Mode was run separately | Confirms the result came from a dedicated Google AI Mode session and that the real answer was captured. |
AIMODE-18 | First question and follow-up | Records the opening question and each follow-up as separate turns, keeping their order and their relationship to the history. |
AIMODE-19 | New brands and sources in follow-ups | Compares, turn by turn, the brands and sources that enter or leave the answer with every follow-up question. |
AIMODE-20 | In-answer links and source panel | Separates inline links in an AI Mode answer from sources that are only listed in the source panel. |
AIMODE-21 | Sub-topic coverage gap | Documents which sub-topics an AI Mode answer covers and which it leaves out. |
AIMODE-22 | Background sub-queries | Records only those background sub-queries that the interface openly shows and leaves the rest marked as unknown. |
AIMODE-23 | Search eligibility checks | Evaluates whether a page is eligible to appear in Google Search, since AI Mode has no separate technical switch for pages. |
AIMODE-24 | Generative AI control scope | Reports your site's generative AI preference for Google as its own record for AI Mode, with its scope. |
AIMODE-25 | Session starting condition | Checks whether an AI Mode measurement can be repeated under known starting conditions. |
AIMODE-26 | Branching plan for follow-ups | Verifies that follow-up questions were planned in advance and that each branch is stored with its history and turn record. |
AIMODE-27 | Source re-matching in every turn | Makes sure each turn's sources are tied only to the answer and the passage they belong to. |
AIMODE-28 | Voice, image and text input | Records whether the input given to AI Mode was voice, image or text, and keeps the evidence for it. |
AIMODE-29 | Experimental or special task modes | Keeps experimental or special task modes of Google AI Mode as separate report series. |
AIMODE-30 | Shopping and local card attribution | Separates the source named on a shopping or local card from the attribution seen in plain text. |
AIMODE-31 | Extra crawling visibility limit | States an information limit for the unseen internal searches AI Mode runs for each question. |
AIMODE-32 | Separation from AI Overviews under equal conditions | Matches AI Overviews and AI Mode results under equal conditions and reports the difference as a separate rate. |
AIMODE-33 | Sub-scope of the Search property check | Shows the owner's AI preferences for AI Mode with the chain of top-level setting and exceptions. |
AIMODE-34 | AI Mode claim limit in Search Console | Prevents data shared by AI Overviews and AI Mode in Search Console from being presented as a separate AI Mode metric. |
The shared contract behind all three
All three surfaces also carry the 30 shared records described on the catalog page. They ensure that, whichever surface is measured, the same discipline applies: the measured product and surface are named, the market and language are recorded, the raw answer and its sources are stored in unalterable form, brands and namesakes are disambiguated, mentions are separated from recommendations, repeats are compared and denominators are shown.
For these three surfaces the shared records have particular importance. ChatGPT’s answers vary between repeats, which is why COMMON-14 separates independent repeats from retries. Google’s AI Overviews appear only sometimes, which is why COMMON-M23 insists on showing the denominator. AI Mode’s conversation depends on history, which is why COMMON-M17 stores the exact conversation version.
What to do with the findings
- Fix access first. If a verified crawler cannot reach a page, nothing else matters until that is solved.
- Make the answer easy to extract. A direct first paragraph, a clear heading and a table with proper headers are easier to cite correctly than a long narrative.
- Remove contradictions. Make prices, addresses and names consistent across language versions, structured data and profiles.
- Use Search Console and analytics properly. Look at breakdowns, but do not add up overlapping totals or treat missing data as zero.
- Re-measure under the same conditions. Improvements only count if the surface, questions and conditions match the earlier run.
For wider context, read the AI search visibility guide. For the three remaining surfaces, see the Gemini, Copilot and Yandex checks. To have these checks run on your site, contact Smartise.
Frequently asked questions about ChatGPT and Google AI checks
Which OpenAI crawler matters for ChatGPT Search?
OAI-SearchBot is the crawler used for search. GPTBot is associated with model training and ChatGPT-User with page fetches made at a person’s request. The audit reports the three roles separately, and the training preference is shown as information only.
Do I need special markup to appear in Google AI Overviews?
According to Google’s published guidance, no special technical requirement exists beyond being indexed and eligible to show a snippet. The audit therefore checks index status, canonical selection and snippet directives, and verifies that structured data matches visible content.
Why might my page not be cited in AI Overviews even though it ranks well?
An overview may not have triggered for that query, or it may have used other sources. Organic rank does not predict AI citation. The audit records whether an overview appeared, then separately whether the page was cited.
Is Google AI Mode data available separately in Search Console?
Search Console data is shared between AI Overviews and AI Mode. The audit does not present it as a separate AI Mode metric, because that would claim precision the data does not provide.
Why does the audit test follow-up questions in AI Mode?
AI Mode is conversational, so the brands and sources in an answer can change with each follow-up. Each turn is recorded separately with its history, so a result is never credited to the wrong turn.
Can ChatGPT visits be measured?
Yes, where you share Google Analytics 4 data. Visits and conversions from ChatGPT are reported separately from citations, because a citation is not a visit.
Continue exploring the AI visibility audit
AI Search Visibility Audit
What the audit is, who it is for, what a report contains and how it differs from a rank tracker.
Read more about AI Search Visibility Audit →397-Check Audit Catalog
How 397 records break down, how to read the numbers and the 30 shared measurement rules.
Read more about 397-Check Audit Catalog →Site Readiness Checks
All 93 site-level checks, grouped by crawl access, indexing, structure, trust and identity.
Read more about Site Readiness Checks →Gemini, Copilot & Yandex Checks
The 66 checks specific to Gemini, Microsoft Copilot and Yandex (Russia).
Read more about Gemini, Copilot & Yandex Checks →AI Search Visibility (GEO) Guide
A practical, hype-free guide to being found, read and cited by AI answer systems.
Read more about AI Search Visibility (GEO) Guide →Ready to grow with Smartise?
Tell us about your business and goals. We will recommend the smallest set of changes that makes the biggest difference, and tell you plainly what we can and cannot promise.