When ChatGPT Counts Your Mentions: Measure AI Traffic Before Business Hub Exists

TL;DR

• OpenAI is testing a ChatGPT dashboard that shows businesses daily impressions and clicks from ChatGPT answers. It sits inside a Business Hub that has not launched yet (Digiday, 9 November 2026).
• The dashboard will not show prompts, user data or competitor ranking. You will get a count, not an explanation.
• Start measuring now: AI referral sessions, AI crawler hits, a weekly prompt panel and the state of your machine-readable surfaces. Then the official number lands on top of a baseline instead of in a vacuum.

What Digiday reported

Digiday’s Krystal Scanlon reports that OpenAI is testing “an analytics dashboard that counts how often a business shows up in ChatGPT’s answers and how many clicks those appearances send to its site.” The feature was spotted by Juozas Kaziukėnas of Marketplace Pulse, who posted a screenshot of a banner asking businesses to “Claim your business in ChatGPT.”

According to Digiday’s sources, the dashboard “logs impressions and clicks each day and plots them on a chart.” It also “stops well short of what search marketers get from Google. Prompts stay hidden. So does user data and there’s no ranking against competitors.”

Claiming a business reportedly also lets you keep product and business information current and unlocks what OpenAI calls a business agent. Proof of authority to represent the business is required; Kaziukėnas mentioned connecting a Shopify store as one route. OpenAI declined to comment for the story.

The commercial context matters too. Digiday notes that ads in ChatGPT went into testing in the U.S. in February, and that OpenAI claimed a $1 billion annualized run rate by September. The article draws the obvious parallel to how Google Business Profile and Search Console made it easier to sell ads to site owners.

Why a count alone will not be enough

I have spent a lot of my career on the gap between a number on a dashboard and a decision someone can make with it. An impressions and clicks chart without prompts tells you that something changed, not why. If your mentions double next month, was it a new article, a product page update, a model change on OpenAI’s side or a seasonal question pattern?

You can only answer that if you have your own series running underneath. The same principle as in how to measure whether AI actually improved anything: record a baseline before the change, measure it the same way afterwards, and compare against something.

Four signals you can measure this week

1. AI referral sessions

When an assistant links to you and someone clicks, the visit usually carries a referrer such as chatgpt.com, perplexity.ai, claude.ai, gemini.google.com or copilot.microsoft.com. ChatGPT links often also carry utm_source=chatgpt.com.

On this site I tag every analytics event with an ai_source property when the referrer or the utm source matches one of those domains. That gives me a weekly count of AI referred sessions per source and per landing page, plus what those visitors did next (scroll depth, time on page, contact clicks). It is a small client script on top of the analytics I already had.

Checklist:

• Classify AI referrers in your analytics, including the utm variant.
• Report sessions per source and per landing page per week.
• Track one downstream action, not only the visit.

2. AI crawler and agent hits

Referrals only show the click. Long before that, crawlers and live fetchers read your pages. Server or CDN logs can count requests from user agents such as OAI-SearchBot, ChatGPT-User, PerplexityBot, Claude-User and ClaudeBot. Search crawlers and on demand fetchers behave differently, so keep them in separate series.

Checklist:

• Decide in robots.txt which AI crawlers you allow, and write it down.
• Count hits per user agent and per path weekly.
• Watch for pages that get fetched often but never get referrals. Those are candidates for better answers or clearer structure.

3. A fixed prompt panel

This is the part the official dashboard will deliberately not give you. Write 20 to 40 questions a real buyer, recruiter or partner would ask, for example “who can help a mid size manufacturer take AI from pilot into production in Sweden”. Run them on a schedule across the assistants that matter to you and record three things per answer: were you mentioned, were you linked, and was the description correct.

Checklist:

• Keep the panel stable for at least a quarter so trends mean something.
• Store raw answers, not only scores, so you can audit later.
• Track correctness, not just presence. Being mentioned with the wrong job title is a bug.

4. Machine-readable surfaces

Assistants and agents read your site in more ways than a browser does. I keep these up to date and treat them as part of the product:

• robots.txt with explicit AI crawler policies and the sitemap.
• llms.txt and llms-full.txt with a short, factual index of who I am and what I have written.
• Person and ProfilePage JSON-LD with a stable @id.
• RSS for new writing.
• A small read-only MCP server at /mcp (documented in llms.txt) so agents can list and read posts as structured data instead of scraping HTML.

There is no public evidence that any one of these raises your mention count on its own. They reduce the chance that an agent gets your facts wrong, and they are cheap. Measure them like any other change.

How to read the official numbers when they arrive

If and when Business Hub ships, put its impressions and clicks next to your own series and ask:

1. Do clicks match my AI referral sessions? Large gaps usually mean tracking issues on one side, not a mystery.
2. Did impressions move when my prompt panel moved? If yes, your panel is a useful proxy for the prompts OpenAI hides.
3. Which pages get crawled, cited and clicked? That shortlist is where to invest writing time.
4. Did anything change after a content release? Mark releases in your series so you can see before and after.

And keep the commercial context in mind. As Digiday puts it, once ChatGPT visibility “shows up as a daily number, marketers will want that number to climb,” and sponsored placement is the lever OpenAI charges for. A baseline of your own makes it easier to tell organic progress from paid lift.

A short checklist to start today

• AI referrers classified in analytics, including utm_source=chatgpt.com
• Weekly report: AI sessions per source and landing page
• AI crawler hits counted per user agent and path
• Prompt panel of 20 to 40 questions, run weekly, raw answers stored
• robots.txt, llms.txt, JSON-LD and sitemap reviewed for accuracy
• Release log so content changes can be lined up against the series

FAQ

What is the ChatGPT Business Hub mention dashboard? According to Digiday, a tested but unlaunched feature that shows businesses daily impressions in ChatGPT answers and clicks to their site. Prompts, user data and competitor rankings are not included.

Can I measure AI assistant traffic without it? Partly. AI referrers, crawler logs and a prompt panel give you a useful picture today.

Why baseline now? Because a first party count without history is hard to interpret. Your own series tells you whether the new number confirms or contradicts what you already see.

Do llms.txt or MCP increase mentions? Not demonstrably. They help agents read you correctly. Treat them as hygiene and measure.

Source: Krystal Scanlon, “ChatGPT is about to show brands their mention count”, Digiday, 9 November 2026.