I Combined Search Console and Bing Webmaster Tools Into a Free GEO Funnel

Google and Bing give you first-party AI visibility data for free. Combine Search Console and Bing Webmaster Tools into a four-question GEO funnel in 30 minutes.

Part of the GEO Roadmap › Stage 7: Measure

You don’t need a paid tool to start measuring AI visibility. Google Search Console’s generative AI report and Bing Webmaster Tools’ AI Performance report give you first-party data for free, and each one fills the other’s gap. Line them up with classic search data and you get a four-question GEO funnel: demand, pull-through, share and value.

Below: what each report shows, the funnel step by step, how my first read of my own data fooled me, what the same funnel revealed on a U.S. SaaS site, and a 30-minute version you can run this week.

There are dozens of paid tools that promise to measure your AI visibility. Most of them do the same thing: they write the prompts, collect the answers and hand you a score. Meanwhile, two search engines now tell you how your site shows up in their own AI, first-hand. Google does it in Search Console’s generative AI report. Bing does it in Webmaster Tools’ AI Performance report. I lined both up with classic search data for the same three months.

The thesis: Each report gives you half the picture

Google shows demand and AI visibility but no queries and no share. Bing shows the queries AI used and your share of citations, but no demand. Combine them and you get a free GEO funnel built on four questions:

1. Am I searched for?
2. When I’m searched for, do I get pulled into AI?
3. When AI answers, how much of the conversation is mine?
4. Does that visibility bring me any value?

I ran it on my own site. My first read fooled me. The second read gave me a much clearer to-do list.

1. Why use both reports?

Here’s who gives you what:

Mike King’s Oct. 15 piece on generative SERPs recommends tracking “retrieved” and “cited” sources separately. These two free reports are the closest first-party version of that split you can get today.

Here’s what I lined up for stradiji.com, July 18 to Oct. 17, 2026: 112,019 classic Google impressions and 508 clicks, plus 6,356 impressions in the generative AI report. In Bing, 215 Copilot citations for the same period. I stripped /tr/, /en/ and trailing slashes so each piece of content is counted once.

2. The funnel: Four questions, three reports

1. Demand: Am I searched for? Source: classic Search Console impressions.
2. Pull-through: When I’m searched for, do I get pulled into AI? Source: generative AI impressions divided by classic impressions. I call it the pull-through rate and read it per 1,000. Next to it sits coverage: what share of your content appears in AI at least once?
3. Share: When AI answers, how much of the conversation is mine? Source: Bing. Divide your citations by your citation share and you get the total citation pie for that query.
4. Value: Does that visibility turn into clicks and sales? Source: Search Console CTR, then GA4 and your CRM.

The headline for stradiji.com: 57 AI impressions for every 1,000 classic impressions, and 36% of 795 pieces of content appear in AI at least once. The real story sits below those averages.

3. Demand and pull-through: How my first table fooled me

Without the funnel, my first read was this: 53.5% of my Google AI impressions go to glossary entries and only 1.9% to my original research. In Bing, research gets 0.5%. I nearly titled the piece “AI cites my glossary, not my research.”

Then I added the first two layers of the funnel:

Research’s small share isn’t about AI. It’s about demand: research gets only 0.8% of classic impressions too. When it is searched for, it’s the type most likely to be pulled into AI, and 11 of 14 pieces appear there at least once.

One caveat: a single piece lifts that average, my study of what AI Overviews did to search in Türkiye. It turns 307 classic impressions into 74 AI impressions. Without it, research drops to 79 per 1,000, still more than double the rate for guides. Fourteen pieces is a small sample. With Mike King’s Sept. 17 warning on precision in mind, I read this as a hypothesis, not a law.

The gap within a content type is even bigger than the gap between types. “What is semantic HTML” pulls through at 614 per 1,000. “What is a SERP” got 1,984 classic impressions and zero in AI. Same template, same author. What decides it is whether the question opens an AI answer at all. Peec AI’s reranker analysis (Aug. 6) says much the same: definitions and lists win when the intent matches, with no blanket format preference.

4. Share: Bing hands you the pie for every query

Bing’s report has a column almost nobody uses: citation share. Of all the sources Copilot cited for that query, this is the share that was yours. Divide your citations by that share and you have the total pie.

Two extremes on my site:

The first is a small pie, and a quarter of it is mine. The second is a big pie, and I’m a crumb. These need different plays. In a small pie, you defend what you own. In a big one, you need to be talked about off your own site to grow your share.

5. Value: Does visibility turn into clicks?

Two things came out of the last layer.

Zero clicks. In six months of classic data (April 19 to Oct. 17, 2026), my glossary pages have a 0.17% CTR against 1.03% for everything else. Copilot cited me 58 times for the Turkish query “chat gpt image creation.” On Google, the same query got 901 impressions in six months, and one click.

Mistaken identity. 56% of my Google AI impressions sit on three pieces: what is a PWA (1,905), ChatGPT image prompts (925) and a list of SEO tools (718). All three are far from what I actually sell, which is SEO and GEO consulting for enterprise brands. To AI, stradiji.com looks more like a PWA resource than a GEO consultancy. Bing’s new Topic Portfolio report (Oct. 13) paints the same picture: for a pure decision query like “ai-driven search visibility expert help,” I’m not there.

The value question is simple: What share of your AI impressions lands on the topic you want to sell?

6. Cross-read 1: Two engines, two different sites

The most unexpected payoff of putting both reports side by side: the two engines read different snapshots of your site.

• Only 16 of the 31 URLs Bing cites appear, with the same URL, in Google’s AI report.
• 70% of Google’s AI impressions sit on the top 10 URLs. Of the six pieces of content behind them, four got zero citations in Bing.
• Bing’s No. 1 is a Turkish “Google search criteria” article I wrote in 2016 and updated on Sept. 27, 2026: 40 citations in Copilot, four impressions in Google’s AI report.
• Duplicate URLs make it worse. The old URL of “What is People Also Ask” got 59 Google impressions; its /en/ version got 6,657 (six months). Bing leans on the old ones: just four legacy URLs carry 24% of my Bing citations.

So “Does AI see me?” has no single answer. Each engine’s AI uses the snapshot of you in its own index. Canonicals and hreflang aren’t a technical footnote here. They’re a precondition for GEO.

7. Cross-read 2: Demand that only exists in AI

Bing’s grounding queries aren’t what the user typed. While building an answer, the model breaks the question apart (query fan-out) and writes its own searches. I matched them against the 8,196 queries in six months of Search Console data.

Bing’s six-month sample has nine grounding queries. Six of them got four or fewer Google impressions over the same six months, and those six carry 31% of my citations. Nobody types “citation optimization meaning techniques” or “ai visibility data content roadmap prioritization” into Google. But that’s exactly what AI searches for while it writes.

I call this hidden demand: demand you never see in Search Console because it only exists in AI’s intermediate steps. Bing’s report is the only free window into it I know of.

8. The language lens: One funnel, two languages

stradiji.com is bilingual, and the funnel shows it. The Turkish ChatGPT image queries are my biggest Bing citations. My Türkiye AI Overviews study is the research piece that lifts my Google pull-through the most. And Bing’s Topic Portfolio barely suggests a single Turkish opportunity query, even though it says the report is personalized to your context.

If you work across languages, run the funnel by language. Local first-party data can travel well into AI once people search for it. The engines’ own suggestions, though, still think in English.

9. Same funnel, giant scale: A U.S. SaaS site

I ran the funnel on a U.S. client site where we have access to both accounts. No name: it’s a SaaS company that offers free browser-based tools and earns from subscriptions. It picked up about 5.9 million Copilot citations in three months.

Volume comes from crumbs of giant pies. What the site really owns is the small brand pie. The value layer agrees: in classic search, brand and core product pages get 3.6% of impressions but the highest CTR, at 7.5%. Download and conversion pages bring the most traffic and rank among the lowest-converting pages.

From Bing’s Topic Portfolio I also built a crowding measure: topic size divided by the number of cited sites. Utility topics get about 10,000 citations per site; the core product category gets ten times fewer. I couldn’t access this client’s Google generative AI report for this study, so the pull-through layer is missing. That’s first on the list for the next round.

10. What the funnel made me do

It gave me three to-do lists.

Technical: We’re consolidating duplicate URLs. The two engines should stop reading different copies of the same content.

Content: Clear answers to questions that open AI answers. Metehan Yesilyurt’s Oct. 7 experiment at Peec AI shows how heavily AI reuses statistics pages (it was built on fake data, so note the result, not the method). Mike King wrote on Oct. 12 that information gain is what gets you cited. My high pull-through research pieces are exactly that kind of content.

Demand: If a piece pulls through well but has low classic impressions, the problem is demand, not content. That’s what our 12-week press release test, which I described in I Asked AI for the Best GEO Agency, is testing. Week 1:

• Coverage: 1,428 placements, for example on USA Today. Ömer Evrey of Pacamara Progress put his experience to work here and made every release, every story, different.
• The real upside: Semrush’s link list (Oct. 12 to 19) also shows sites that saw the story and wrote about it themselves. Mike Gingerich, founder of Digital Hill, summarized the study on his blog and walked through the three stages of our GEO Score Card.
• Numbers: 190 new referring domains. Our share in GeoGenie’s press release prompts is still zero. According to Notified’s study, 56% of such citations arrive after week one.
• Its trace in the funnel: If the release creates demand, the research page’s classic impressions should rise first, then its AI impressions. I’ll show both in week 6 (around Nov. 23).

For the value layer: as I wrote when HubSpot lost 140 million visits and still posted record revenue, most lost traffic was never going to buy. We connected stradiji.com to a free HubSpot account through HubSpot’s MCP connection in Claude, so form leads land there directly. Measurement has just started; I’ll report results in six weeks.

The one thing to do this week: Build your own funnel

It takes about 30 minutes.

1. Export Search Console’s generative AI report and classic performance report by page, for the same three months. From Bing, export AI Performance’s page and grounding query lists.
2. Normalize URLs (language prefixes, www, trailing slash) and merge them.
3. Pull-through and coverage: Divide each page’s AI impressions by its classic impressions and multiply by 1,000. What share of each content type appears in AI at least once?
4. Share: In Bing, divide each grounding query’s citations by its citation share. Which pies do you own, and where are you a crumb?
5. Cross-read: Do the two engines pick the same pages? How many of your grounding queries never show up in Search Console?

If a piece has a high pull-through rate but low classic impressions, don’t write more of it. Get more people talking about it.

Closing

I combined two free reports and my first read fooled me. A piece saying “AI doesn’t read my research” nearly went out. Adding the demand layer flipped the table. Adding the share layer showed me which pies I own. Adding the cross-read showed me the two engines see me differently.

GEO is a new field. Nobody knows everything here, and neither do I. What I do is set up an assumption, measure it, and drop it if it doesn’t hold. If you meet someone who claims to know it all, or who says they’ve fully automated SEO, ask to see the data before you book the second meeting.

Which layer of your funnel leaks the most: demand, pull-through, share or value? If you’d like a second pair of eyes on your numbers, this funnel is where our GEO consulting work starts.