Search Console, Bing and your own prompts: what each AI visibility report counts

Three reports, three different counts

Google, Microsoft and prompt-based monitoring each answer a different question. Search Console tells you how often your pages were shown inside Google's AI features. Bing tells you how often Microsoft's AI answers cited your pages. A prompt monitor tells you what a set of chosen questions returned across engines, including the ones neither platform reports on.

The temptation is to merge them into one "AI visibility" figure. Resist it. An impression, a citation and a mention are different events with different denominators. Adding them produces a number that moves for reasons nobody can trace.

This post sets out what each report counts, where each goes blind and how to read them together each month. It assumes you already have access to Search Console and Bing Webmaster Tools. If you do not, set both up before you buy any tool, because they are free and the data is first-party.

What each report counts

Search Console AI reports

• Impressions in AI Overviews, AI Mode and Discover AI features
• By page, country, date and device
• No clicks in the AI reports
• Google surfaces only

Bing AI Performance

• Citations and cited pages per day
• A sample of grounding queries
• Copilot, Bing AI summaries, some partners
• No clicks

Prompt-based monitoring

• Mentions, citations and accuracy per prompt
• ChatGPT, Claude, Perplexity and others
• Your prompts, not real search volume
• Varies run to run, so repeat runs

What does Search Console's AI report count?

Impressions, and only impressions. As of 31 August 2026, Search Console reports impressions from AI Overviews, AI Mode and generative AI features in Discover for all sites, broken down by page, country and date, with device for Search.¹ The reports do not include click data.¹ Sites without enough AI impressions may not see them yet.

Where the clicks went

Clicks from AI features have not disappeared; they sit inside the ordinary Performance report. Google's guidance says traffic from AI features is included in the overall figures under the Web search type.³ So you can see that a page was shown in AI Mode, and you can see the page's total clicks, but not how many of those clicks came from the AI feature.

What an impression means here

An impression means your page appeared within an AI feature on a results page. It does not mean the answer quoted you, recommended you or named your brand. Search Engine Journal's report puts it plainly: the reports show where pages appear, not how many clicks those appearances generate.¹

Our view: treat AI impressions as a reach figure for Google's AI surfaces, and nothing more. Rising impressions can mean your pages are cited more often. They can also mean Google is showing AI features on more of the queries you already ranked for. The report alone cannot tell those apart.

The same day, Google added a property-level setting to keep a site's content out of these AI surfaces. Google says the setting is not used as a ranking signal elsewhere in Search.¹ Our note on the rollout and the opt-out covers whether anyone should use it.

What does Bing's AI Performance report count?

Citations. Bing's AI Performance report, in public preview since February 2026, counts how often your pages are shown as sources in AI-generated answers.² It covers Microsoft Copilot, AI-generated summaries in Bing and select partner integrations, which Microsoft does not name.²

It has four views. Total citations over a period. Average cited pages, meaning unique pages cited per day. Grounding queries, the phrases the AI used to retrieve your content. And citation activity per URL over time.²

Grounding queries are not search queries

Grounding queries are the most useful view and the easiest to misread. They are the phrases the AI system used when it went looking for sources, not what a person typed. A user's long conversational question may become several short retrieval phrases. Microsoft also says the data shown represents a sample of overall citation activity.²

Read them as evidence of which topics your pages are trusted for. If grounding queries cluster on one product line and skip another, that tells you where your content is retrievable. It does not tell you what share of all Copilot answers in your category cite you.

What Bing leaves out

No clicks, and no engines outside Microsoft's surfaces and partners. The preview does not say that ChatGPT is covered, so do not assume it is.² Treat any claim to the contrary as unconfirmed until Microsoft names the partners.

It is also a preview. Metrics, coverage and definitions can change before general release, so note the date of every export and keep the raw files. If a definition changes mid-quarter, a saved export is the only way to compare like with like. The same habit applies to Search Console, whose AI reports were still rolling out to some properties at the end of August.

Why the platform reports are not enough

Because the engines your buyers use are not only Google's and Microsoft's. Google's surfaces are huge: Alphabet said in July 2026 that AI Mode had passed 1 billion monthly active users and that the Gemini app had 950 million.⁴ But a large share of AI questions happen in tools that publish no report for site owners.

The market is also spreading out. Similarweb estimates that ChatGPT's share of worldwide generative AI web traffic fell from roughly 76% a year earlier to around 53%, in data published in July 2026.⁵ In the US, ChatGPT prompts carrying a citation rose from about 1.6% in June 2025 to about 6.8% in May 2026.⁵ Neither Google nor Bing can show you any of that.

Why one report is not enough

Big audiences, outside every webmaster tool

1B+monthly active users of Google's AI ModeAlphabet, Jul 2026 ⁴

About 53%ChatGPT's share of worldwide gen-AI web traffic, down from about 76%Similarweb, Jul 2026 ⁵

About 6.8%of US ChatGPT prompts carry a citationSimilarweb, May 2026 ⁵

The low citation rate matters for measurement. If ChatGPT names your brand without linking anywhere, no referral appears in analytics and no platform report records it. The only way to see it is to ask the question and read the answer.

What does prompt-based monitoring count?

Answers to questions you choose. A monitor sends a fixed set of prompts to each engine on a schedule and records what comes back: whether your brand is mentioned, where, whether a page of yours is cited, which competitors appear and whether the facts are right. It covers engines that publish nothing for site owners.

Its blind spots

The prompts are a sample you picked, not the questions people actually typed. There is no search volume attached, so a prompt monitor cannot tell you how many people saw an answer. And answers vary between runs, so a single run can mislead.

Those limits are why method matters more than the dashboard. Our AI answer monitor queries six engines: ChatGPT, Gemini, Perplexity, Claude, Copilot and Google AI. It uses official APIs where the engine offers them and a licensed AI-answer data provider otherwise. Our audits run each prompt on three separate days, because one day's answer is an anecdote.

Our view: the prompt set is the measurement. Build it from questions your sales and support teams hear, in the countries you sell to, and freeze it for at least a quarter. Change the prompts every month and you are measuring your own edits. Our buyer's spec for AI visibility scores sets out what a defensible prompt set and margin of error look like.

How do impressions, citations and mentions relate?

Loosely, and in one direction. A citation in an AI answer usually produces an impression in the platform that showed it. A mention in a ChatGPT answer produces neither, unless it links. And an impression in Search Console can happen without your brand being named in the text at all.

Sources: Search Engine Journal; Bing Webmaster Blog; Google Search Central. "Not stated": Microsoft does not name its partner integrations.

The link between being cited and getting clicks is real where it has been measured. In Seer Interactive's study of 53 brands (January 2025 to February 2026), pages cited in an AI Overview earned about 120% more organic clicks per impression than pages that were not cited.⁶ In the same data, organic click-through on queries with an AI Overview rose from 1.3% in December 2025 to 2.4% in February 2026.⁶

Organic click-through on queries with an AI Overview

February 20262.4%⁶

December 20251.3%⁶

That is a correlation in one vendor's client base, and Seer sells search services. It still gives the monthly read-out its logic: impressions are the reach, citations are the share of that reach that names you, and clicks are what finance will ask about.

A combined monthly read-out

One page, five rows, a written note. Each source keeps its own row and its own unit, and the note explains movement in words. Here is the order we use.

Build the monthly read-out

1. Export Search Console

   AI impressions by page for the month, plus total Web clicks for the same pages.

2. Export Bing

   Total citations, average cited pages and the top grounding queries.

3. Run the prompt set

   Same prompts, same engines, on three separate days; record mention and citation rates.

4. Pull AI referrals

   Sessions and conversions from AI engines in your analytics, with the channel rules unchanged.

5. Write the note

   What moved, which source shows it, and the likely cause, in three sentences.

Illustrative. Northwind Digital is fictional and every figure is an invented round number.

The note for that month might read: AI impressions rose by about a quarter while cited prompts were flat. Most of the new impressions came from two comparison pages, which suggests Google showed AI features on more of our existing queries. ChatGPT now names us on three more prompts, without linking.

Our view: the note is worth more than the table. It forces someone to explain the numbers, and it stops a rising impression count being presented as a win when nothing else moved. For the analytics row, our guide to seeing AI referrals in analytics covers the channel rules.

Who should own each row?

One person per source, and one person who writes the note. The Search Console and Bing rows usually belong to whoever owns SEO, because they already live in those tools and know which pages changed. The prompt-set row belongs to whoever owns the prompt list, since every change to it has to be logged. The analytics row belongs to the person who set the channel rules.

The note needs a different owner: someone senior enough to say what the numbers mean for the business and to ask why when they disagree. In a small team that may be the same person throughout. That is fine, as long as the prompt set and the channel rules are written down where someone else can check them.

Our view: date every change to the measurement itself, such as a new prompt, a new engine or a changed channel rule, in a line under the table. Six months later, that log is the only way to tell a real shift from a change in method.

Common misreadings to avoid

Four mistakes come up again and again. Each one turns a real number into a wrong conclusion.

• Reading rising AI impressions as rising preference. They may just mean Google shows AI features on more of your queries.
• Treating Bing's grounding queries as keyword volume. They are retrieval phrases, and a sample.
• Comparing a prompt monitor's rate between vendors. Different prompts, engines and run counts produce different rates.
• Adding the three sources into one score. The units do not add, and the total moves for reasons nobody can trace.

A fifth is subtler: changing the prompt set and the reporting period at once. Then nobody can say whether visibility moved or the measurement did. Our explainer on tracking AI Overviews and AI Mode separately covers a related split inside Google's own surfaces.

What belongs in the board report?

Two numbers and a sentence, not the whole read-out. Pick one reach figure, such as Google AI impressions on your money pages, and one outcome figure, such as AI referral conversions. Add one sentence on share of answer from the prompt set.

Keep the full read-out for the marketing team. Our board-report guide covers which AI search number belongs at that level and how to label it so nobody mistakes impressions for customers.

Google says AI features in Search send billions of clicks to websites every week.⁴ That is a company statement about all sites, not a measure of yours. Your read-out is how you find your share of it.

Where to start this month

Start with the free, first-party data. Open Search Console's AI reports and note which pages carry AI impressions. Open Bing's AI Performance and read the grounding queries for those same pages. If the two lists disagree, you have learned something about how each engine sees your site.

Then write down the twenty questions your buyers ask most and run them in ChatGPT, Gemini and Perplexity yourself, on two different days. It is slow, but it shows you the gap the platform reports leave. When doing it by hand stops being practical, that is the point to automate it. Our guide to how often to re-run checks covers the schedule.