How this page was sourced — every platform fact was read on July 14, 2026 from Google's and Microsoft's own documentation, plus the IndexNow spec. Where the free rails stop, our own pass is measured the same way we ask you to read theirs: 7 repeat runs per prompt-surface pair, reported with a Wilson 95% interval. That interval is the whole reason a repeat pass beats a screenshot, because one reading of a 95% band is a guess about where the real rate sits and seven readings narrow it.
Do you need a paid tool to see your AI search visibility? Start with the rails you already own, then buy what they cannot reach. Two official reports are free: Search Console's generative AI report and Bing Webmaster Tools' AI Performance preview. GA4's AI-referred sessions add a third rail of your own data. This page sets out what each rail proves and what it cannot reach. The Bing rail counts double, because its index feeds Yahoo Search results under a long-standing deal and sits under Microsoft Copilot. Where the official rails stop, panel measurement carries on. Our Panorama diagnostic runs the full grid with your rivals on the same axes, and reports rates with Wilson score 95% confidence intervals.
Which official AI-visibility reports already exist?
Most teams shopping for an "AI visibility" dashboard have not yet opened the reports the platforms publish themselves. Google Search Console now ships a generative AI report of its own. It is rolling out property by property. It covers impressions from AI Overviews and AI Mode, and you can read it by page, country, device and date, then export it from the screen. Official primary support.google.com
On the Microsoft side, Bing Webmaster Tools announced AI Performance as a public preview on February 10, 2026. It reports citation activity across Copilot, Bing's AI-generated summaries, and select partner integrations: total citations, average cited pages per day, citation trends, sampled grounding query phrases, and citation counts per URL. Official primary blogs.bing.com
Both come from the operator of the surface. Both are tied to your verified property, and both cost nothing. If AI search matters to your pipeline, turn these rails on before you start looking at tools. They are the floor a vendor's numbers have to line up with. And what they cannot show you is exactly what you should be paying for. Where these two reports sit among the answer surfaces themselves is drawn out in our US AI search surface map.
What does Google's generative AI report prove, and what can it never reach?
The report answers one question, and it answers it with the platform's own authority. How often did links to your site show up inside Google's AI experiences, and where? The counting rules are published. A click on an outside link from AI Mode counts as a click. A follow-up question counts as a new query. An AI Overview link counts as seen once the part of the answer holding it is opened or scrolled into view, and links in the same box share one position. Official primary support.google.com
Five things it never carries. The prompts people typed. The answer text. The exact URL that was cited. What the answer said about your brand. And whether the citation earned a click. As of our July 14, 2026 review we found no public read API for this report, so plan around screen exports rather than a feed. One habit matters from day one. If the report has not appeared on your property yet, that is a rollout state, not proof of zero AI impressions. A blank and a zero are different facts. Official primary support.google.com
What does Bing's AI Performance preview prove, and where does it stop?
Bing's preview is the closest thing to official citation evidence on the market today: the platform itself telling you how often your pages were used as grounding, with sampled query phrases and per-URL counts. That is a real signal you can watch move. Official primary blogs.bing.com
Read it for what it is. A citation count proves your pages are being pulled into answers. It does not prove clicks. The preview also holds back the full answer the reader saw, where your citation sat inside it, and what it did next. It is a screen, not a feed: as of our July 14, 2026 review we found no public read API, so treat exports and captures as the record. Rising citations mean your content is in the grounding pool. Which questions, said how, against whom: that lives outside this report.
One more reason the Bing rail punches above Bing's own share, noted July 17, 2026. Bing is not one surface. Its index feeds Yahoo Search results under a long-standing deal, sits under Microsoft Copilot, and, by several industry analyses, still shapes which sources ChatGPT search puts up. Market data searchengineland.com
What does GA4 tell you, and what do submission receipts not tell you?
The third rail is your own analytics. GA4's session source and medium fields show sessions sent over by AI services. That is an honest floor for AI-driven visits, and a floor is all it is. A referral only shows up when the service sends a referrer, and apps, browsers and redirect rules quietly file many of those visits under direct or unknown. Read it as the hard count of arrivals, not as the size of AI's pull. Official primary developers.google.com
Your raw server log sits under that rail and costs nothing either. We measured our own log across the fourteen days to 2026-07-30: ten AI-family crawlers came back to the same pages 447 times, with a median gap of one day. That tells you how quickly a fix on a page can even be seen by the systems that build answers, which is a question no dashboard on this page answers. It is a crawler count, not a citation count, and the two should never be added together.
One trap to clear while you wire this up. Indexing and submission APIs measure nothing. Google's Indexing API is limited by policy to job postings and livestream markup. A success response from any submission channel, IndexNow included, means the request arrived. It does not mean anything was crawled, indexed, ranked or cited. Keep delivery receipts in a different drawer from performance evidence. Official primary developers.google.com · indexnow.org
Does an IndexNow success response mean you were indexed?
No. A success response from any submission channel — IndexNow included — means the request was received, nothing more: it is not evidence that anything was crawled, indexed, ranked, or cited. File submission receipts under delivery, keep performance evidence on the reporting rails above, and never let one stand in for the other.
Why keep the three rails separate?
Everything above sorts into three rails, and the operating rule is simple: report them side by side, never blended into one score.
The habit that makes this work is stating, for every cell in a report, whether it was observed, reported as zero, available only in a UI, or simply not offered by any official surface. Zero and unobserved are different answers to different questions, and a structure that refuses to confuse them is what lets you compare a Google number, a Bing number and a panel number without fooling yourself.
The panel rail applies the same discipline at run level: a run with no AI answer shown is recorded as its own state, kept out of the citation denominator and published as a count, instead of being booked as a zero for the brand. This is the same rail-and-state frame our own reporting runs on — the full framework is public on our methodology page.
What can none of the official rails show you?
Line the rails up against the chain you actually care about and the gap is visible.
No official rail tells you which buyer questions your brand shows up in. None tells you what the answer says when it does. None tells you whether the citation beside your category goes to you or to a rival. Those are the facts that decide where your next content and PR dollar goes. They are also what a controlled panel is built to catch: a fixed set of real buyer questions, asked again and again across the AI surfaces that matter, with every appearance, mention and citation logged per question. And what the controlled research says actually earns those citations is collected in our evidence review of what gets cited.
CiteAngle runs that panel as Rail C, built to the same evidence rules as the official rails beside it. Our Panorama diagnostic runs the grid in full and shows the status of every cell. Rates that carry one are reported with Wilson score 95% confidence intervals, so you can tell a real move from noise before you shift budget. You get the official rails' story and the inside of the answers, in one report that never mixes the two.
How do you stand the rails up this week?
Five steps, in order. Each one is free, and each one ends with something you can put in a folder. Copy this checklist into your own tracker and date every line as you finish it.
- Verify the property. Search Console and Bing Webmaster Tools both need a verified site before any report appears. Do this first, because the reports only start counting from verification.
- Open the generative AI report in Search Console. If it is not there yet, write down the date you looked. A blank is a rollout state, not a zero.
- Open AI Performance in Bing Webmaster Tools. Export the citation counts per URL and the sampled query phrases. Keep the export, not a screenshot of a number.
- Split AI referrals out in GA4. Build one report on session source and medium, and label it a floor. It counts arrivals that carried a referrer, and nothing else.
- Write down what is still missing. Name the buyer questions you care about, and note that no rail above tells you what the answer said about you on any of them. That list is your scope for anything you buy next.
Every step points at a report the platform documents publicly, linked in the sections above and checked July 14, 2026. Nothing here needs a vendor.
What buying blind costs. Without the free rails standing first, you risk paying a vendor to answer a question the platform was already answering. Where those reports stop, our roster reads 15 engines on the same questions, sealed with their conditions and with competitors read side by side with you — which is the part no platform report can reach.
The reason to use us for the rest is a difference you can check for yourself. Readings are sealed at measurement time, with the questions, the engines, the run count and the judging rule attached. A figure you quote then survives an audit. Competitors are read side by side with your brand inside the same answers, which no platform report will ever do for you.
Why we tell you to start with the free rails
It is a strange thing for a paid vendor to lead with, and it is deliberate. A buying team that stands up the official reports first arrives at any vendor conversation knowing what their own data already says, which makes them much harder to sell to on vibes. We would rather meet that team. It also protects you from the ordinary failure in this market, where a tool gets bought to answer a question the platform was already answering for free.
Where those rails end is where we begin, and the boundary is easy to state. The official reports tell you about traffic that reached you. They cannot tell you what the answer said, which brands were named instead of yours, or which sources the engine cited to build it — and no platform report covers the engines outside its own house. We read 15 of them on the same questions, with your competitors inside the same answers and every reading sealed with its conditions. Bring the free rails, and use us for the part they structurally cannot see.
Stand up the official rails — then measure what they can't show you
Search Console and Bing report your own impressions. The cross-engine read — who gets named and cited when your buyers ask — is what the measured grid adds, with receipts.