Bing Topics Portfolio: The First Keyword Planner for AI Agents?

On October 13, 2026, Microsoft introduced Topics Portfolio (Beta) inside the AI Performance report in Bing Webmaster Tools. I've been exploring the data, and I keep coming back to one question.

Are we looking at the first keyword planner for AI agents?

Not quite. But it's the closest thing any search engine has shipped so far, and it's worth understanding both what it shows and where it misleads.

First, the Good News on That Chart

Here's the Citations & Pages view for one of the sites where we've been running optimization loops around AI fan-out queries. Over the last three months it has collected 4.3M citations in Bing's AI ecosystem, with an average of 212 cited pages, and both lines are climbing.

Nice to see the upward trend. 🌴

But the more interesting part is the new banner under the chart: "Find where to grow in AI Search. Discover the topics driving citations to your site and where the biggest gaps are." That's the door into Topics Portfolio.

What Topics Portfolio Actually Shows

Bing describes a topic as "a broad thematic cluster of related grounding queries where your site appears." The landing table lists your topics with four columns:

Open a topic and you get three cards (Topic Size, Your Citations, Queries) and two tables side by side. Your Grounding Queries lists the top queries where your site already appears, with citation counts. Opportunity Queries lists the top queries in the same topic where other sites are cited and yours isn't, with citation volumes shown as ranges. You can view one, two, or three months.

That's a genuinely new kind of data. Search Console and Bing's own Search Performance report tell you what people typed. This tells you what the AI typed when it went looking for sources, and who it picked.

The Reality Check: The Big Numbers Mislead

Look at the topic above:

• 100M to 500M estimated citations across the ecosystem
• 500K to 1M associated queries
• 219.3K citations to our site

Impressive numbers. Then you scroll through the actual "opportunities."

Many are generic, extremely broad, or completely disconnected from the business. You can find topics with enormous estimated reach that have practically zero relevance to the client's products, audience, or commercial goals. A topic that big is mostly made of queries no one would ever want to rank for commercially.

A large topic doesn't mean a large business opportunity. And calling every gap an "opportunity" is a bit generous. :)

Bing is upfront about the limits, which I appreciate. In its own words, topic classification "is AI-generated and can occasionally be imperfect," numbers are approximate and "deliberately shown as ranges," branded queries "can appear as opportunity queries," and smaller sites may see "a shorter list or may not even see this feature."

Pro tip

Before you share a Topic Size number with anyone, open the topic and read twenty of its opportunity queries. If you can't connect at least a handful of them to something the business sells or knows, the topic size is a vanity number for you, however large it is.

Where It Gets Interesting: Below the Topic Level

The value isn't in the topic sizes. It's one layer down, in the grounding queries themselves, once you do the work Bing doesn't do for you:

1. Export the queries: both views have a Download button. Pull your grounding queries and the opportunity queries for every topic that's even plausibly relevant.
2. Classify them by intent: informational, comparison, transactional, troubleshooting, branded. Grounding queries read differently from human searches, so intent has to be judged per query, not assumed from the topic.
3. Filter out the irrelevant: drop the generic, the off-topic, and the competitors' brand queries (Bing warns these show up as "opportunities").
4. Map what's left to the business: which product answers it? Which page should be cited for it? Is there real expertise behind it, or would you be writing to a query?

In practice, we take the Topics Portfolio export, cluster the grounding and opportunity queries into tighter topics than Bing's, and add up what Bing reports for each one: your citations from Your Grounding Queries, and the citation ranges on the Opportunity Queries that are going to other sites. Then we score every cluster on two sides Bing can't see: how close it sits to what the business sells, and how well an existing page already answers it. That turns a long export into a short list of decisions. Each cluster becomes a new page, an optimization of a page that half-answers it, a topic that's already covered, or a skip.

Look at the two gray bubbles on the left. They hold the most citations going to other sites by far, and they are both skips. Caffeine and bean-origin questions are exactly the kind of huge, generic demand a topic size would make look like a priority, and almost none of it leads to someone buying an espresso machine.

What comes out the other end usually isn't the biggest queries. Sometimes it's quite the opposite: narrow, specific, high-intent queries with modest citation ranges that sit squarely on what the client sells.

Pro tip

Run the same classification on Your Grounding Queries, not just the opportunity list. The queries you're already cited for show you what the AI already trusts you on, and the closest neighbors of those queries are usually the cheapest gaps to close.

A New Input for Zero-Volume Keyword Research

A few days ago I joined a great discussion started by Dan Petrovic about finding high-value, converting queries that traditional keyword research tools miss because they report little or no search volume.

This dataset could become another valuable input into that research. Think about it:

• Traditional keyword research tells us what people search for.
• AI Performance is starting to show us which queries AI systems use to ground their answers, and which sources they cite.

That's a different layer of demand intelligence. Plenty of grounding queries will never show up in a keyword tool because no human types them that way. The AI rewrites the user's question into its own fan-out queries, and those queries shape which pages get retrieved. That's potentially a new way to find content gaps that conventional keyword research would never surface.

Of course, the caveats matter:

• Citations aren't conversions: this report counts appearances, not visits, just like the AI reports Google and Bing launched earlier this year.
• Grounding queries aren't necessarily user prompts: they're what the system searched for, which is related to, but not the same as, what the person asked.
• Bing's AI ecosystem isn't the whole market: it covers Microsoft Copilots and partners, not ChatGPT, Gemini, or Google's AI surfaces.

So is this keyword research for AI agents? I think it's the early stages of it. Not a replacement for traditional keyword research, but an additional dataset that helps us understand what AI systems retrieve, reference, and use to construct answers.

Why This Matters for Our Clients

For the sites we work on, this changes the research step, not the strategy.

We already track how clients appear across ChatGPT, Gemini, Google AI Overviews, and AI Mode in our LLM visibility monitoring, and we already build content around the fan-out queries those systems generate. Topics Portfolio adds a first-party view of Bing's side of that: real grounding queries, real citation counts for our own pages, and a list of queries where competitors are being picked instead.

What we won't do is take the topic sizes at face value or chase every "opportunity" because it has a big range next to it. The work is the same as it's always been with a new data source: classify it, filter it, and connect the right queries to actual business outcomes.

For now, I'd treat the big numbers with caution and the underlying queries with curiosity.

Anyone else already exploring this feature? I'd love to hear what you're finding beyond the obvious generic terms.