ChatGPT in Google Search Console: What You Can Track

Google Search Console does not track ChatGPT.

It counts clicks and impressions from Google Search. That's all. A visit from chatgpt.com never shows up in it.

But it does show two AI signals:

1. The Generative AI report. Impressions in AI Overviews and AI Mode. All sites have it since 31 August 2026.
2. AI Mode follow-ups. Someone asks a follow-up in AI Mode. Google counts it as a new query. It lands in your normal Performance report.

Here's what each report shows. Plus the regex filters I teach, and how often they missed on this site's own data.

What Search Console can and cannot show about AI

Search Console reports on Google Search.

Google's definition of a click: a user clicking through to your site from Google Search results.

So a visit that starts in ChatGPT, Perplexity or Copilot is not a Search Console click.

Google's own AI features are the exception. AI Overviews and AI Mode count toward your overall search data. They sit in the Performance report, inside the Web search type.

Came to connect Search Console to ChatGPT? Jump to that section.

Came for AI Mode? Read the next two sections.

The Generative AI report for AI Overviews and AI Mode

Google announced its Search Generative AI performance reports on 3 June 2026. All websites have them since 31 August 2026.

The Search report covers AI Overviews and AI Mode. Discover has its own.

What you get:

• Impressions. How many times links to your site were shown in a generative AI feature.
• Four groupings. Pages, Countries, Devices or Dates.
• A search type filter. Web: text-based, or Web: multimodal (image-led).
• Export. The usual 1,000-row limit applies. (Google's help page.)

What you don't get:

• A Queries tab. So the report can't say which search triggered an AI feature.
• A split. AI Overviews and AI Mode are counted together.
• API access. Search Engine Journal's Suganthan Mohanadasan wrote on 19 August 2026 that he re-checked his own property on 11 August. Neither the Search Analytics API nor the BigQuery bulk export exposes this data.

So the report is UI only.

Can't see the report? Google gives two reasons:

1. Google's help page still lists the rollout as a reason. Its own note says all sites have the report since 31 August 2026.
2. Your site hasn't had enough impressions in generative AI features.

Use it to see which pages Google shows in AI features. For the query side: how to track Google AI Overviews.

AI Mode queries sit in your Performance report

There's no AI Mode filter in Search Console.

Google counts AI Mode in the same Performance report as everything else. Its counting rules decide what you see:

• A click on an external link in AI Mode counts as a click.
• Impressions follow the standard rules.
• A follow-up question is a new query. The impressions, position and clicks of the new response go to that new query.
• An AI Overview takes one position. Every link in it gets that position.

That has a funny side effect.

In August 2026, Anastasia Kourou shared queries from her Performance report. Things like:

• yes
• yes go on
• yes, pricing

Nobody searches Google for "yes go on".

John Mueller replied with Google's counting page. Barry Schwartz wrote that we have to assume such queries came from AI Mode.

It's an assumption. Google doesn't label them.

So "how do I track AI Mode in GSC" has two parts:

1. The Generative AI report: which pages get AI impressions.
2. The Performance report: follow-up fragments and long, prompt-like queries.

Neither tells you which engine, or which person, sent a query.

Find AI queries with regex filters

No setting separates AI queries from human ones.

So the workaround is pattern matching.

Metehan Yesilyurt published the method in June 2025. He calls it a workaround. Not an official tracking solution.

(Barry Schwartz credited the first word-count regexes of June 2025 to Metehan, Vijay Chauhan and David Konitzny. Search Engine Journal repeats his credit. Thanks to StoryChief for the data in the first screenshot.)

How to apply a filter:

1. Open the Performance report.
2. Click Add filter.
3. Choose Queries.
4. Choose Custom (regex).

Google uses RE2 syntax. Matching is partial unless you anchor it with ^ and $. It ignores case by default.

I teach this in my AI SEO course. Students keep asking for one ultimate regex.

There is none. But these three help.

(I checked them on sample strings first. The test on this page comes later.)

75 or more characters:

.{75,}

10 or more words:

^(\S+\s+){9,}\S+$

A month followed by a year. The year avoids the verb in "may i":

\b(january|february|march|april|may|june|july|august|september|october|november|december)\s+20\d\d\b

Metehan also suggests the plain words reply, task and research.

An earlier version of this post taught "latest" and the year too. I dropped both. StoryChief's screenshots for them showed mostly ordinary queries.

Example 1: the word "task"

StoryChief filtered for "task".

Seven of the ten rows are long, instruction-style strings. Like this one:

"you are an seo expert tasked with analyzing a brand and creating comprehensive keyword lists"

Each has 12 to 14 impressions. No clicks. Of the three short queries, one got 3 clicks.

Real people don't type prompts like that. I read these as AI tools running research.

Search Console can't prove it. But Quantable reported in October 2025 that ChatGPT prompts leaked into Search Console as queries. OpenAI claimed it fixed the leak and said it touched a small number of prompts.

Example 2: a month and a year

This is from a 1:1 consulting session with our Pro plan user tradervue.com.

A month-name regex over three months. It matched queries with 200 impressions and 0 clicks.

The top one: "weekly options trading strategies may 2025". 92 impressions.

I read that as AI-style research.

Careful though. A month and year also match ordinary searches, like "march 2026 core update". A hit is a lead. Not proof.

Example 3: 75 or more characters

Another client, voicenotes.com. The 75-character regex found conversational questions. Like:

"how can i integrate miro with my existing project management and communication tools?"

Zero clicks on 32 impressions for that query. That's not a click-through problem. I read it as AI-style research that sends no traffic.

My view, not a measured fact: one long query replaces three to five classic searches. Things like "best crm" and "crm pricing".

How well the length rules work

I tested them. On this very page.

Search Console lists 30 queries for this page, 1 July to 30 September 2026. 366 impressions. 0 clicks.

Five of the 30 read like prompts:

• "how do i combine google search console data with ai engine citation tracking to understand where my traffic is really coming from?" (130 characters, 22 words)
• "what tools show the performance of our content in ai-driven search results?" (75 characters, 12 words)
• "how can i discover which questions users ask ai in my category?" (63 characters, 12 words)
• "how to check if i have traffic from search gpt or perplexity" (60 characters, 12 words)
• "how to measure ai referral traffic to a news site" (49 characters, 10 words)

The 10-word rule won.

But it's one small sample. And a person can type ten words.

All five got 0 clicks. Search Console can't show if a person, a voice assistant or an AI tool typed them.

So use the rules to build a shortlist. Then read each query.

Three limits on every filter here:

• Hidden queries. Google leaves out anonymized queries: the ones that a few dozen users or fewer issue over two to three months. Prompt-like strings are rare. So they're the ones most likely to be hidden. Search Engine Journal's author measured 57.7% of his own impressions in that pool over 59 days. Treat any AI-query list as an undercount.
• Row cap. The UI shows and exports at most 1,000 rows.
• English. The pattern lists are built for English.

$49 report

Does AI name you?

See if AI mentions your brand. Then get the pages to fix.

Your website

Get my $49 report

Check one question freeHow it works

What's the best AI SEO tracker for a startup?

Most answers name Semrush. You aren't cited.

Ask ChatGPT

Where ChatGPT visits actually show up

In your analytics. Not in Search Console.

The UTM tag.

OpenAI says publishers who allow OAI-SearchBot can track referral traffic from ChatGPT in tools like Google Analytics. And that ChatGPT adds utm_source=chatgpt.com to referral URLs in its search results. (OpenAI publishers FAQ.)

So filter your landing pages by that source.

The GA4 channel.

GA4 has a default AI Assistant channel. It's for users who arrive from sources like ChatGPT, Gemini, Deepseek, Copilot or Grok.

It excludes Google's AI Overviews and AI Mode.

The page doesn't name Perplexity or Claude. Look for them in your referral sources.

The one place Search Console shows ChatGPT.

In March 2026, Glenn Gabe reported that chatgpt.com appeared among the top linking sites for some client sites.

Those are shared ChatGPT conversations that cite the site.

Where to look: Links, then External links, then Top linking sites.

• Not every site has them.
• Gabe says to check directory properties too.
• He doesn't know if the links will stay.
• Google says the Links report is a sample, not a full list.

Copilot has its own report.

AI Performance in Bing Webmaster Tools shows citations of your site across Microsoft Copilot, AI summaries in Bing and select partner integrations. Citation counts per page, plus a sample of grounding queries.

Microsoft does not name the partners. ChatGPT is not listed.

Connect Search Console to ChatGPT

Many readers come here for this: give ChatGPT your Search Console data.

There's no official connector. (Windsor.ai, a connector vendor, says in its FAQ that neither Google nor OpenAI publishes one.)

Three routes exist. Windsor.ai and Supermetrics are vendors. Examples, not endorsements.

A connector sees more rows. Google's 2022 guide says the API returns up to 50,000 rows per day per site per search type. Against 1,000 for the UI export.

But ChatGPT can only analyze what Search Console exposes.

• The Generative AI report is not in the API, according to Search Engine Journal's author.
• Anonymized queries are missing.

So a connector can't find the AI queries that Search Console hides.

Check a new connection. Windsor.ai's tip: ask for clicks and impressions for a period. Compare them with the Performance report.

Then try this prompt on an export of your queries:

Here is a list of Search Console queries with impressions. Flag queries that read like pasted prompts, replies such as yes go on, or agent instructions. Return the query and a one-line reason.

Pair Search Console with AI answer tracking

Search Console shows what people ask Google. And which pages Google shows.

It doesn't show what ChatGPT, Perplexity, Gemini or Copilot say about your brand.

So put three sources side by side:

1. Search Console: Google demand and AI impressions.
2. Analytics: AI referrals.
3. Answer tracking: mentions and citations.

In the tradervue example above, our client built content for how AI searches. It now gets more traffic from AI engines.

AI SEO Tracker measures the answers. $49, once.

It scans ChatGPT, Perplexity, Gemini, Copilot and Google AI Overviews. Daily for 7 days. One brand. Up to 50 prompts, written from your site. You confirm them.

You get mentions, citations, position in the answer, competitor gaps and the sources each engine cited.

It doesn't scan Google AI Mode, Claude, Grok, DeepSeek or Meta AI.

Tip: when you confirm prompts, use the real phrasing you found in Search Console as a check.

The full method: how to measure AI brand visibility and AI search monitoring tools.