AI search visibility in 2026 cannot be measured with a single metric or dashboard. A useful reporting framework combines crawler access, presence inside AI answers, referred sessions, and business outcomes across Google, ChatGPT, Perplexity, and other AI search surfaces.
Key takeaways:
- AI visibility should be separated into four layers: crawl access, presence in AI answers, referred sessions, and conversions
- Google Search Console now provides dedicated generative AI impression reporting for AI Overviews and AI Mode, but does not expose query, click, CTR, or citation-position data in that report
- ChatGPT visibility needs to be measured through crawler logs, referral traffic, and consistent manual prompt testing because OpenAI does not provide a webmaster analytics console
- OAI-SearchBot, GPTBot, ChatGPT-User, and OAI-AdsBot serve different purposes, so they should be tracked separately rather than grouped into one AI crawler metric
- Perplexity can be measured through server logs, analytics referrals, fixed prompt sampling, and manual tracking of citation position
- A fixed monthly set of commercially relevant queries helps measure citation presence and competitor visibility where platform dashboards provide no direct data
For about two years, the honest answer to “are we showing up in ChatGPT?” was a shrug. Agencies said yes because it sounded good, clients believed it because they had no way to check, and nobody could put a number on it that survived a follow-up question.
That changed in 2026, but only partly, and the part that changed is easy to overstate. Google now reports AI visibility inside Search Console. ChatGPT and Perplexity still do not. So the practical job is no longer waiting for a dashboard. It is building a measurement approach that mixes the data you can now get with the data you will have to collect yourself, and being clear with clients about which is which.
This guide sets out what each platform actually exposes, how to capture the rest, and how to assemble it into reporting that does not fall apart the next time a vendor changes something.
1. First, separate four things people call “visibility”
Most confused reporting comes from collapsing four different states into one word. Keep them apart and the measurement design writes itself.
Crawl access is whether the AI platform’s crawler is permitted to fetch your pages at all. This is binary, it lives in your robots.txt and your server logs, and it is the only layer you control completely.
Eligibility and presence is whether your pages appear inside an AI answer. This is what Google now reports as impressions, and what you have to sample manually everywhere else.
Referred sessions is whether a human clicked through to your site from that answer. This lives in your analytics and it is typically a fraction of presence.
Business outcome is whether those sessions did anything. Enquiries, signups, purchases.
A report that shows only referred sessions understates your position badly, because most AI answers resolve without a click. A report that shows only impressions overstates it, because an impression is not a visit. The whole point of the exercise is that these are different numbers, and the gap between them is the story. This is the same tension that made zero-click search so difficult to report on, now applied to a wider set of surfaces.
2. What Google now gives you
On 3 June 2026, Google announced Search Console’s generative AI reports, a dedicated view of how often your pages appear inside Google’s generative AI features. The reports cover AI Overviews and AI Mode on Search, with a separate report for generative AI features in Discover. Google initially released them to a subset of sites, and as of 31 August 2026 has rolled them out to all websites worldwide.
Two details matter for how you use this.
First, the data is not new. AI visibility was already being counted inside the overall Performance report, mixed into the Web search type with everything else. What the dedicated view adds is the ability to isolate it. That means you must not add the AI numbers to your existing Search totals. They are already in there.
Second, the dedicated view is built around impressions. It breaks them down by page, country, device, and date. According to Google’s help documentation, click, click-through rate, and query data are not part of this report, and there are specific reasons a site might see nothing at all: the property has not received enough AI impressions, or the site has been excluded from generative AI features.
So what you can now say confidently to a client is which of their pages are being surfaced inside Google’s AI answers, and whether that is rising or falling. What you still cannot say from Search Console alone is what people were searching for when it happened, or how many of them came through.
Practical advice: set a baseline now, mark the pages that carry AI visibility, and annotate every significant content or technical change against the chart. Historical data does not backfill, so the value of this report compounds from the day you start watching it. If you want the background on how these answers are assembled in the first place, our explainer on how AI Overviews work covers the mechanics.
3. What Search Console still will not tell you
Be honest about the gaps, because clients will find them.
You cannot see the query behind an AI impression, which means you cannot tell whether you appeared for a commercial question or an informational one. You cannot see your citation position inside the answer, or whether you were the primary source or a supporting link. You cannot see the answer text, so you cannot tell whether the AI summarised you accurately or contradicted you.
All three require manual sampling, and there is no way around it. Build a fixed set of twenty to forty queries that matter commercially, run them under consistent conditions once a month, and record whether an AI answer triggered, whether the client was cited, and who else was. It is tedious and it is the only source of that information.
4. Measuring ChatGPT
OpenAI provides no webmaster console. What it does provide is clear documentation of its crawlers, which turns your own server logs into a usable measurement surface.
OpenAI’s crawler documentation lists four user agents, and the distinction between them is the whole basis of sensible log analysis.
OAI-SearchBot is the search crawler. It is what surfaces sites in ChatGPT’s search features, and OpenAI states that sites opted out of it will not be shown in ChatGPT search answers, though they can still appear as navigational links. If you want to be citable in ChatGPT, this is the one that must be allowed. Note that robots.txt changes can take around 24 hours to take effect for search.
GPTBot crawls content that may be used to train OpenAI’s foundation models. Blocking it is a separate decision with separate consequences, and it does not block search inclusion.
ChatGPT-User is used when a person asks ChatGPT or a custom GPT to visit a page. It is not automatic crawling, and because the action is user-initiated, robots.txt rules may not apply. OpenAI is explicit that this agent is not what determines whether content appears in Search.
OAI-AdsBot visits pages submitted as ads on ChatGPT to check them against policy.
Read as measurement signals, these tell you different things. OAI-SearchBot activity tells you your content is being indexed for ChatGPT search. ChatGPT-User activity is closer to a demand signal, because a real person prompted that fetch. Tracking the two separately over time is considerably more informative than a single “AI bot hits” figure, and it lets you spot the case where you are being crawled heavily but never actually retrieved for users.
Alongside logs, two other signals are worth capturing. Referral traffic from ChatGPT appears in analytics, and ChatGPT tags outbound links with a source parameter that makes it identifiable, so you can build a channel for it rather than leaving it buried in Referral or Direct. And prompt testing, the same fixed-set discipline described above, applied to the assistant rather than the SERP. Our guide to ranking in ChatGPT search covers the optimisation side of this.
One caution worth passing on to clients. Referrer data from AI surfaces is incomplete by nature, since many assistants render answers without a normal click path. Treat referral numbers as a floor, not a total.
5. Measuring Perplexity
Perplexity sits in the same position ChatGPT does: no official webmaster analytics, so you assemble the picture from your own data.
The approach is identical in shape. Identify its crawler user agents from the vendor’s current published list rather than from a blog post, since these change and a stale allowlist quietly costs you visibility. Track those agents in your logs. Segment referral sessions in analytics. And run the same fixed prompt set you use for ChatGPT so the two are comparable.
The one thing worth doing differently is watching citation behaviour rather than just presence. Perplexity displays sources prominently, which means position in the source list is visible in a way it is not inside a Google AI Overview. If you are sampling manually anyway, record the position.
6. A framework that survives the next platform change
The tooling here will keep moving. Build your reporting around the four layers rather than around any one vendor’s dashboard, and it will still work in a year.
Layer one: access. Which AI crawlers can reach your content, verified in logs rather than assumed from robots.txt. Report this as a checklist, not a trend.
Layer two: presence. Search Console AI impressions for Google, manual prompt sampling for everything else. Report as a trend plus a sampled citation rate.
Layer three: referred sessions. A dedicated channel in your analytics covering AI assistant referrers, so the traffic is visible rather than absorbed into Direct. If your analytics setup is not currently capable of this, our guide to tracking your website traffic covers the groundwork.
Layer four: outcome. Conversions from those sessions, measured the same way you measure conversions from any other channel.
The discipline is reporting all four together. Layer two without layer three lets an agency claim success with no commercial evidence. Layer three without layer two makes AI look irrelevant because the volumes are small. The combination is what actually informs a decision. The same principle applies here as in traditional reporting, which we set out in our piece on measuring SEO effectiveness.
7. What the measurement should change in your work
Measuring is only useful if it feeds back into what you produce.
If your pages are being crawled but rarely cited, the problem is usually the content rather than the access. AI systems tend to quote material that states things plainly, attributes claims, and answers a question in a self-contained passage rather than across five paragraphs of preamble. The structural habits that support this overlap heavily with writing a genuinely useful page, which is why our SEO-friendly blog post process still holds up.
If you are cited but the citations misrepresent you, the fix is clarity and evidence on the page itself. Named authors, dates, sources for statistics, and unambiguous statements of what you do and for whom. The E-E-A-T content guide covers the signals that matter here.
And if you are neither crawled nor cited, start with access before rewriting anything.
8. Four mistakes worth avoiding
Reporting a single vendor “AI visibility score” as fact. Third-party tools estimate presence by running prompts at scale. That is a useful sample, not a census, and different tools produce different numbers for the same site. If you use one, report it as an index and say so.
Comparing across a reporting change. Google’s dedicated reports only carry data from the point the feature began collecting it, and the rollout reached all sites at the end of August 2026. Any year-on-year comparison that spans that boundary is not a like-for-like comparison.
Treating impressions as traffic. They are not, and a client who is told otherwise once will not trust the next number either.
Blocking crawlers you have not looked up. The difference between the search crawler and the training crawler is the difference between staying citable and disappearing. Make that decision deliberately, with the vendor’s own documentation open, not by copying a robots.txt from a forum post.
Final thoughts
AI search measurement in 2026 is in an awkward middle state. One platform reports impressions but not clicks or queries. The others report nothing and leave you to reconstruct the picture from logs, referrers, and manual sampling. Anyone selling you a complete, automated view of AI visibility is selling an estimate with a confident interface.
That is not a reason to skip it. The businesses that start measuring now build a baseline that nobody can retrofit later, and they find out which of their pages actually carry AI visibility while there is still time to do something about it. The ones that wait for a tidy dashboard will get one eventually, and it will start counting from the day they turn it on.
The practical version is simple enough to start this month. Confirm which AI crawlers can reach your content. Turn on and baseline the Search Console AI reports. Build a segment for AI assistant referrals in your analytics. Write down twenty commercial queries and check them by hand. Then do the same thing again in thirty days. Everything more sophisticated than that is built on those four habits, and none of it works without them. The wider shift behind all of this is something we covered in our look at SEO 2.0, and the direction of travel has not changed.
If you would rather have this set up properly and reported monthly than build it in-house, take a look at our SEO services and we will put a measurement framework in place that fits how your business actually makes money.
Frequently Asked Questions
What is the difference between AI search visibility and AI referral traffic?
AI search visibility means your content appears or is cited inside an AI-generated answer, while referral traffic only counts people who actually click through to your website. The two should be measured separately because many AI interactions end without a website visit. A site can therefore gain significant visibility while receiving relatively little referral traffic from the same platform.
Can Google Search Console show which queries triggered an AI Overview or AI Mode citation?
Not from the dedicated generative AI reporting described in this article. Search Console can show AI impressions and break them down by dimensions such as page, country, device and date, but it does not provide the underlying query or the exact citation position. Commercial query visibility therefore still needs to be assessed through a consistent manual sampling process.
How can I measure whether my website appears in ChatGPT if there is no webmaster dashboard?
Use several signals together rather than relying on one number. Server logs can show activity from OpenAI's different crawlers, analytics can identify referral sessions from ChatGPT, and a fixed set of commercially important prompts can be tested regularly to record whether your site is cited. These signals measure different stages, so crawler activity should not be treated as proof that your content is actually appearing in answers.
Does being crawled by OAI-SearchBot mean my website is visible in ChatGPT answers?
No. Crawler activity confirms that ChatGPT's search crawler can access and retrieve your content, but it does not prove that the content is being selected for user-facing answers. That is why access data should be reported separately from actual presence or citation data gathered through prompt sampling.
How should manual AI visibility testing be carried out consistently?
Build a fixed set of commercially relevant queries and run the same set at regular intervals, such as once a month. Record whether an AI answer appears, whether your brand or page is cited, which competitors or other sources appear, and where useful, the citation position. Keeping the prompt set and reporting method consistent makes month-to-month movement more meaningful than occasional spot checks.
Are third-party AI visibility scores reliable?
They can be useful as directional indicators, but they should be treated as sampled estimates rather than complete measurements of platform-wide visibility. Most tools generate a predefined set of prompts and calculate an index from what they observe, so different vendors can produce different scores for the same website. The most defensible reporting combines these estimates with first-party data such as Search Console, server logs, analytics and your own repeatable prompt testing.
