AI answers have been quietly skipping your business. Now you can see it.

A homeowner three neighbourhoods over asks an assistant who to call for a burst pipe. It names three plumbers in a sentence. For two years, you had no way to know whether you were one of them.

That is the part that has been genuinely maddening about “AI search.” Everyone agrees it matters. Every tool on the market will sell you a “track your AI rankings” dashboard. And the search engines themselves quietly admitted the underlying data did not exist in any form you could check. You were told to prepare for the future with no instrument pointed at the present.

What this has felt like from your side

We hear the same things from established service businesses — roofers, HVAC companies, dentists, law firms, electricians — almost word for word:

• “We rank #1 on the map and the AI still names someone else.” A screenshot of your Google Maps position tells you nothing about whether an assistant pulled your page into its answer.
• “I’ve done everything right for a decade and a newer competitor keeps getting mentioned.” Usually because their information is easier for a machine to verify, not because their work is better.
• “Calls are down and I can’t tell if it’s the season, the economy, or AI quietly rerouting my customers.” Without a citation number, that is a guess every time.
• “My agency pitched ‘AI visibility’ and there was no metric behind it.” Hard to approve a line item you cannot verify.
• “We publish good content and the AI summarises it for free while the click never arrives.” Real problem. The report below is the first step to measuring it instead of resenting it.

None of that is solved by a clever trick. But the not-knowing part — that just changed.

The blindfold came off: Bing’s AI Performance report

In February 2026, Microsoft added an AI Performance section to Bing Webmaster Tools, in public preview. It is the first time a search engine has published a report on how its AI systems cite your site. Bing’s own framing of why it exists:

“As AI becomes a more common way people discover information, visibility is not only about blue links. It is also about whether your content is cited and referenced when AI systems generate answers.”

The report covers Copilot, AI-generated answers in Bing, and select partner integrations. That last part matters more than it sounds: several assistants people actually use lean on Bing’s index to fetch and ground their answers, so a Bing-side citation report is a usable window into more than just Bing itself.

What it shows, in the plainest terms we can put it:

• Total citations. How many times your site was listed as a source in an AI-generated answer over the period you pick. This is the headline number — the one you can finally point at in a meeting.
• Average cited pages. The daily average number of individual URLs from your site that showed up as sources. Tells you whether one page is carrying you or your whole site is pulling weight.
• Grounding queries. A sample of the actual phrases the AI was answering when it reached for your page — the closest thing yet to seeing the question behind the citation.
• Page-level citation activity. Citation counts per URL, so you can see which pages get cited and go make more like them.

You reach it at bing.com/webmasters/aiperformance once your site is verified. Two honest limits up front: the data is a sample, not every citation that ever happened, and it counts citations, not visits. It is a visibility signal, not a traffic report.

What Bing says actually earns the citation

The same announcement lists what Bing recommends to get cited more often. Read it next to Google’s AI-search guidance from last July and it is almost the same list — which is the point. Translated out of documentation-speak:

• Depth and real expertise on the pages that already get cited. Not more pages. Better ones, written by someone who has actually done the work.
• Clear structure — headings, tables, a real FAQ. So a machine can lift a clean, self-contained answer without guessing.
• Back your claims with examples, numbers and attribution. “We’re the best” is not groundable. “Here is the data, here is the source” is.
• Keep the information current. Stale hours, old service lists and dead pages all read as a business that may not still operate the way the page says.
• Consistent facts across text, images and video. The same story everywhere an assistant might check.

Notice what is not on that list: a special file, a robotic writing style, chopped-up content, a pile of low-quality mentions. Both Google and Bing have now said, in their own words, that a business already doing normal SEO well does not need a separate bolted-on AI strategy. If that sounds familiar, it is the same argument we made when Google wrote its AI search rulebook.

The mistakes that don’t show up until you check

The detector covers the ones people argue about. These are the quieter ones — the kind that never announce themselves, because nothing on your site looks broken:

• Letting your NAP drift. A suite number on Google, no suite on Yelp, an old tracking number on an industry directory. Every mismatch is a reason for an assistant to trust your data less.
• Thin service-area pages. A page per town with the name swapped, no author, no specifics, no evidence. There is nothing there for an AI to ground on, so it grounds on a competitor.
• Chasing mentions on any site that will have you. Assistants building local answers lean on the directories and review sites customers actually check, not a scattershot of low-quality links.
• Treating a rank-tracker screenshot as proof. A rank checker cannot see inside an AI answer. It never could. The citation report can.
• Never opening the reports that would tell you the truth. Bing’s AI Performance report and Search Console’s AI view both exist now. An unread dashboard is the same as no dashboard.

Your best proof is in your head. That’s the leak.

Here is the one that costs established businesses the most. Your crews finish real jobs in real neighbourhoods every week. That is exactly the specific, first-hand, verifiable evidence an assistant wants — and almost none of it ever reaches a page. It lives on a phone, in a text thread, in your memory of a good week. Google knows you work there; your website doesn’t, and neither does the AI.

Getting that proof onto the site in a dated, mapped, machine-readable form is the mechanism behind Map Check-Ins: finished jobs, timestamped and pinned, published to your site and Google Business Profile as they happen. It is not an AI trick. It is just making the evidence exist where a machine can read it.

Reading the number without fooling yourself

A citation report is powerful and easy to over-read. A few rules we hold ourselves to:

• It is a sample. A month with more citations than last month is a real signal. A single day up or down is noise.
• Citations are not calls. Being cited is discovery. Whether it turns into a booked job still depends on the website behind the click and how the phone gets answered.
• Grounding queries are a gift, not a keyword list. They tell you what customers actually ask an assistant. Write for those questions like a human, not by stuffing the phrasing back into a page.
• One number, checked over time, beats five dashboards checked once. Pick the metric — total citations for your key pages — and watch it across a quarter.

This is the same standard we apply to map visibility: a real baseline, re-measured on a schedule, compared honestly — including the months it does not move.

What you actually want out of this

Strip away the anxiety and the pitch, and the goal is short:

• When someone asks an assistant for the best in your trade and city, your name is one of the ones it says.
• A report you can open that shows, in a number, whether that is happening — and whether it moved this quarter.
• One set of fundamentals that works for Google, Maps, AI Overviews, ChatGPT and Copilot at once, not five projects and five invoices.
• The AI describing you as the local expert, because the evidence on your site earns that word.
• Confidence you are already built for the next AI surface before it arrives, because the foundation is right.

The honest part

This is early. The report is a public preview, the data is sampled, and Google still has not shipped a local-business equivalent that is as direct. Both companies are also talking about “agentic” assistants that will browse a site and book the appointment themselves — real, coming, and not something we would sell you a fix for today. What we can say honestly is that the work that gets you cited now is the same work that has always earned trust: consistent facts, genuine expertise, and proof a stranger — human or machine — can check. We would rather build that than a product for a surface that does not exist yet. More on how we handle AEO and GEO →