Bing Webmaster Tools AI Performance Topics: What to Do
Bing's AI Performance report now sorts the queries where your site was cited into topics, and it lists queries where other sites are cited and yours isn't. For AEO and GEO teams, that is the first first-party view of citation gaps from a major search engine, and it is worth using as a planning input. It is not a scoreboard to optimise against line by line.
Key takeaways
- Bing's AI Performance report now groups cited queries into topics and lists queries where other sites are cited but yours isn't, according to Search Engine Journal.
- Topics shift AEO work from single queries to subject areas, which matches how content and product teams already plan.
- The citation-gap list is a research queue. Check intent and your existing coverage before you commission anything.
- Measure per topic over time, with a baseline for each, because AI answers vary from one run to the next.
- Don't chase every gap, don't rewrite pages to mimic a competitor, and don't treat one engine's data as the whole of AI search.
What Bing added to the AI Performance report
Bing Webmaster Tools already had an AI Performance report that shows where your pages are cited in Bing's AI-generated answers. Search Engine Journal reports that the report now has two additions. Cited queries are sorted into topics, so you can read performance as a set of subjects instead of a flat list of phrases. And there is a list of queries where other sites are cited but yours isn't.
That second part is the one most teams will open first. Until now, finding out who was being cited instead of you meant running prompts by hand, or paying for a third-party tracker that sampled answers on your behalf. Both approaches work, but both are sampling. A report from the engine itself is a different kind of evidence, even if it only covers that engine.
What the report does not tell you is just as important. It covers Bing. It won't describe what ChatGPT, Gemini, Perplexity or Google's own AI features do with the same questions. Bing's index does feed some other assistants, so the signal is not irrelevant outside Bing, but treat that as a hypothesis for your own testing rather than a rule. Check Bing's documentation and your own dashboard for exact metric definitions, date ranges and how topics are assigned.
Why topics matter more than queries for AEO
A single query in an AI answer is a noisy unit. The same question phrased two ways can pull different sources. The same phrase run twice can too. If your team reports on individual prompts, you spend a lot of meetings arguing about whether a drop is real.
Topics are steadier. A topic like 'invoice financing for small exporters' contains dozens of phrasings, and your presence across them says something about whether the engine considers your site a reasonable source for the subject. That is closer to how a person on a content team thinks: do we cover this area well, or not?
It also gives you a cleaner way to split work. A topic maps to a page cluster, a product line or a market. Someone owns it. Compare that with a list of 400 queries, which belongs to nobody.
There is a caveat. Someone else's grouping may not match yours. Bing's topics are built from how the engine clusters queries, and your site architecture was built around your products and funnel. Expect some topics to straddle two of your sections and some of your sections to split across several topics. Make a simple mapping sheet once and keep it; that mapping is more valuable than any single report export.
How AI answer engines choose what to cite
Nobody outside the vendors knows the exact selection logic, and it differs between engines. A few mechanics are well enough understood to plan around.
First, an answer is usually assembled from several retrieval steps, not one. Ahrefs published a study of how ChatGPT expands a prompt into multiple searches, which is a good illustration of the general pattern: one user question becomes a handful of narrower lookups, and the sources are drawn from all of them. That is why a site can be cited for a sub-question the user never typed.
Second, passages matter more than pages. An engine tends to lift a clear, self-contained answer to a specific question. A page that buries the answer under a long introduction, or spreads it across tabs and accordions, gives the engine less to work with.
Third, retrieval still depends on ordinary search plumbing: the page has to be crawlable, indexed, reasonably fresh where the topic demands it, and not blocked from the bot in question. Many citation gaps are not content problems at all. They are indexing problems.
Fourth, the engine needs a reason to trust the source for that subject. Clear authorship, a consistent entity description, specific claims that can be checked and a track record on the topic all help. None of this is new. What is new is having a report that shows where it isn't working.
Turning the citation-gap list into a work queue
The list of queries where others are cited and you aren't is tempting to treat as a ticket backlog. Don't. It is a list of places to look. Each row has one of a few possible causes, and the fix depends on which.
- Export and group. Pull the gap list and sort it by topic. Ignore individual phrasings at first and look at which topics have the most gaps.
- Check whether you have a page at all. Search your own site for the subject. If nothing answers the question, that is a coverage gap.
- If a page exists, check Bing's view of it. Is it indexed? Is the canonical what you expect? Is the relevant content in the HTML or injected late by script?
- Read the cited competitor page. Not to copy it. Ask what it does that yours doesn't: a direct definition, a worked example, a table, a recent date, a named author with credentials.
- Decide the smallest change that closes the gap. Often it is a clearer opening paragraph or a missing section, not a new article.
- Log the change and the date, so you can attribute any movement later.
Here is how the causes usually sort out in practice:
| Likely cause | How to confirm | Typical fix | Owner |
|---|---|---|---|
| No page covers the question | Site search and sitemap review | New page or new section on an existing page | Content lead |
| Page exists but isn't indexed or is canonicalised away | Bing URL inspection in Webmaster Tools | Fix canonical, internal links, crawl blocks | SEO or engineering |
| Page covers it, but the answer is buried | Read the first screen as a stranger would | Lead with a direct answer, add a short summary or table | Editor |
| Page is thin next to the cited one | Side-by-side comparison of specifics | Add examples, data you own, named authorship | Subject expert |
| Answer is stale | Check dates, product names, prices | Refresh and show a real updated date | Content owner |
Notice that only two of those five fixes involve writing something new. Teams that skip the diagnosis step tend to produce more content and see the same gaps.
How to measure progress without fooling yourself
AI answers are not stable. A topic-level view helps, but you still need discipline in how you read it.
Set a baseline per topic before you change anything. Record the number of cited queries, the number of gap queries, and the date range. Then change a small set of pages and leave the rest alone for long enough that you can see a difference. Bing's documentation will tell you how often the report refreshes; plan reviews around that, not around how impatient the stakeholders are.
Keep a control. If you refresh pages in one topic and leave a comparable topic untouched, you have some way to tell whether movement came from your edits or from the engine changing its behaviour. It won't be perfect. It is much better than nothing.
Track three things together, not one:
- Citations by topic from the Bing report.
- Organic clicks and impressions for the same pages from standard search reports, since AI citation and ordinary ranking often move together and sometimes don't.
- Business outcomes: leads, sign-ups or assisted conversions from the landing pages involved.
The third is the one that stops the programme drifting into vanity. A page cited often but producing no enquiries is telling you something about the audience or the offer, not about AEO.
If you want cross-engine visibility rather than Bing alone, that is where a dedicated tool earns its place. MediaPilot is Cresia's product for media and AI-search visibility; see the product page for what it covers.
Who should own what
The report is only useful if someone is accountable for acting on it. A plain split works for most mid-size teams:
| Task | Owner | Cadence |
|---|---|---|
| Review topic report and gap list | SEO lead | Fortnightly or monthly, matched to report refresh |
| Map Bing topics to site sections | SEO lead with content lead | Once, then update when site structure changes |
| Decide which gaps are worth closing | Content lead | After each review |
| Technical checks: indexing, canonicals, rendering | Engineering or technical SEO | As gaps are flagged |
| Tag landing pages and track outcomes | Analytics lead | Ongoing |
| Annotate changes for later attribution | Whoever ships the change | Every change |
The analytics row deserves more attention than it usually gets. If page changes aren't annotated and your tracking isn't consistent, you can't tell what worked. A written tracking specification keeps event and campaign naming stable while pages get reworked, and teams that want to run this as a governed process can look at how analytics teams set it up.
What not to do with this data
Some patterns look productive and aren't.
Don't chase every gap. Some queries where competitors are cited are ones you shouldn't answer. They may be outside your expertise, off-strategy, or served better by a page type you don't want to build. A gap is only a problem if the query matters to your business.
Don't rewrite pages to echo a competitor's wording. You may close a gap and lose what made your page distinct. Engines are choosing between sources; being a near-duplicate gives them no reason to pick you.
Don't generate dozens of thin pages, one per gap query. It is the fastest way to produce a site that is large and unhelpful. One strong page per topic area usually beats ten near-identical ones.
Don't treat Bing as the whole market. If most of your audience arrives through Google or uses other assistants, use this report as one input among several. A fix that makes a page clearer will help everywhere, but don't tune details to one engine's quirks.
Don't report citations as revenue. Present them as a leading indicator with an honest caveat, alongside outcomes you can actually tie to the pages.
What this signals about AI search measurement
The interesting part of this update is less the feature than the direction. Search engines are starting to give publishers first-party reporting on their AI surfaces, after a period in which the only option was sampling prompts yourself. Expect more of this, expect it to be uneven between vendors, and expect definitions to change. Reports may be revised, topics regrouped, metrics renamed.
So build your process around things that survive those changes: a topic map you control, a baseline per topic, a log of what you changed, and outcome tracking on the landing pages. If a vendor's report is replaced next year, you lose a data source, not the programme.
If your marketing operations are still scattered, it helps to settle ownership first; what marketing operations covers is a useful starting point for that conversation.
Frequently asked questions
Does the Bing AI Performance report cover ChatGPT or Google AI Overviews?
No. It reports on Bing's own AI-generated answers, as described by Search Engine Journal. Other assistants may draw on Bing's index to some degree, but you should verify that with your own testing rather than assume it.
Should we write new content for every query where competitors are cited?
No. First check whether you already have a page, whether it's indexed, and whether the query is worth answering for your business. Many gaps close with a clearer opening paragraph or a technical fix, and some aren't worth closing.
How often should we review the topic report?
Match the cadence to how often the report refreshes, which you can confirm in Bing's documentation or your dashboard. For most teams a fortnightly or monthly review is enough. Checking daily mostly measures noise.
How do we know if a change actually improved citations?
Keep a baseline per topic, change a limited set of pages, annotate the date, and compare against a similar topic you left alone. It won't give you proof, but it will keep you from crediting yourself for movement that came from elsewhere.
Is this worth doing if most of our traffic comes from Google?
Yes, as a diagnostic. Clear answers, working indexing and strong authorship help across engines. Just don't make Bing-specific tweaks the centre of the plan.
Sources
- https://www.searchenginejournal.com/bing-webmaster-tools-topics-portfolio-ai-performance/592874/
- https://ahrefs.com/blog/query-fan-out-chatgpt/
- https://www.conductor.com/academy/aeo-geo-benchmarks-report/