I measured whether AI actually cites my site. Here is the method, and the uncomfortable result.

Four months ago I published an AI visibility audit of this site and opened it by complaining that most AI SEO content is theory. Reading it back, mine was too. It scored whether the site was ready to be cited: schema, llms.txt, an AI bot allowlist. It never checked whether anything actually cited us.

It turns out you can check, for free, and hardly anyone does. Here are the real numbers from this site, including the part that undermines the sales pitch.

Readiness is not measurement

Almost every AI visibility audit on sale right now scores inputs. Is your schema valid. Do you have an llms.txt. Are the AI crawlers allowed. Those things matter, and they are also entirely within your control, which is precisely why they are easy to sell and easy to score well on.

None of it tells you whether a single AI answer has ever mentioned you. That is an outcome, it sits outside your control, and until recently it was genuinely hard to see. It is not hard any more.

The report almost nobody opens

Bing Webmaster Tools now carries an AI Performance tab, in beta, alongside the familiar Search Performance report. Verifying a site is free and takes minutes. The report gives you four things:

• Total citations over your chosen window, with a daily series
• Citation share, the proportion of answers to a given query that cite you
• Grounding queries, the actual questions that produced a citation
• Cited pages, so you can see which URLs are doing the work

The citation share figure is the interesting one. Ranking tells you where you sit in a list. Citation share tells you how often you make it into the answer at all, which is a different and increasingly more relevant question.

What it showed for this site

Three months, 2 June to 31 August 2026. This is a small site with one referring domain, so treat the absolute numbers as a floor rather than a benchmark.

129 citations against 139 search impressions. Roughly one to one. Half of this site's visibility inside Microsoft's ecosystem was happening in a channel that appears in no standard analytics report, and I had no idea until I looked.

Five things the data said

1. Two pages did almost all of it

Of 129 citations, 119 came from two blog posts. That is 92%.

If you are commissioning AI visibility work, this is the shape to expect. It is not a site-wide property. A small number of pages carry it, and knowing which ones changes what you do next.

2. AI cited the teaching, not the selling

Every meaningful citation landed on an informational post. The pricing page picked up three. The service pages, the ones that exist to be bought from, earned zero. If your AEO plan is to get your services page cited, the data does not support it.

3. A geo-scoped page won a query with no geography in it

The two queries producing citations were "how to choose the right SEO agency for business" and "how to choose the right SEO agency". Neither mentions a country. The page being cited was written for one specific market, and it won anyway, because it was the closest thing available.

That is a gap worth acting on. If a geo-scoped page can win a global question, a page actually written for the global question should do better.

4. Citation share can be high while everything else is small

26.78% on the best query means roughly one in four answers to that question referenced this site. That is a strong number, produced by a site with almost no backlinks and three clicks a quarter from Bing search. Citation share and traditional authority are not the same currency.

5. It moves in ways you cannot yet explain

The monthly split was 2 in June, 111 in July, 16 in August. That looks like a collapse. It might be one. It might also be sampling, because Bing labels the data "a sample of overall activity". One quarter is not enough to know, and anyone telling you they can read a trend from a beta report after three months is guessing.

The uncomfortable part

129 citations produced no enquiries I can trace. Not one.

A citation is not a visit. Someone asked Copilot how to choose an SEO agency, Copilot answered using this site, and the person got what they needed without arriving here. That is the zero-click problem with a new name, and it is worse than the search version because there is no impression, no position and no click to report on.

So citation share is a real metric and a poor goal. It measures whether a machine trusts your page enough to repeat it. It says nothing about whether a human then does anything. Anyone selling AEO on citation share alone is selling you a number that does not have to convert into anything.

What it is genuinely good for: finding out which of your pages the models actually trust, which is not the set you would guess, and which questions they trust you on. That is a content strategy input. It is not a lead source.

Run it on your own site

Fifteen minutes, no cost.

1. Verify your site at Bing Webmaster Tools if you have not already. You can import from Google Search Console.
2. Open AI Performance in the left navigation. Set the window to 3M or 6M, because 7 days will show you almost nothing.
3. Read Total Citations first, then compare it to Search Performance impressions over the same window. The ratio is the interesting part.
4. Switch List By to Grounding Queries and note the citation share column. These are the questions the models associate you with.
5. Switch to Pages. Expect concentration. Ask why those pages and not the ones you assumed.
6. Check whether any commercial page appears. If none does, that is your gap.

What this does not measure

Be precise about the scope, because the temptation to overclaim is strong. The report says its sources are Microsoft Copilot and partners. That is not ChatGPT, not Perplexity, and not Google AI Overviews. Those systems publish nothing comparable today.

The data is also sampled and the feature is beta, so figures may be revised. Treat it as a directional instrument on one ecosystem, not a full picture of AI visibility. It is still infinitely more than the zero measurement most audits are built on.

Related

• The readiness audit this post corrects
• Which sources AI engines cite on recovery questions
• The Google Update Calendar, free dataset of every confirmed update