How do you actually check whether ChatGPT and Google’s AI are recommending your business?
Everyone selling “AI visibility” right now is selling a dashboard. Nobody agrees on what the number in it means. Here is what you can actually measure today, what it costs, and which parts of the pitch Google itself contradicts.
Quick answer if you’re skimming
- Google Search Console now has a generative AI report, free, live on all sites since 31 August 2026 — but it shows impressions only, not clicks.
- There is no API or log that tells you why ChatGPT or Gemini named someone. Every tracking tool is sampling prompts and guessing.
- Google’s own documentation says there are no special requirements to appear in AI Overviews or AI Mode, and names llms.txt as a tactic you can ignore.
- The traffic at stake is real but smaller than the panic suggests: with an AI summary on the page, people clicked a normal result 8% of the time versus 15% without.
- A manual prompt log in a spreadsheet, 20 prompts checked monthly, gets you 80% of what a paid tracker gets you. Start there.
Is there anything in a tool I already pay for?
- Yes. Search Console’s generative AI performance report rolled out to all sites worldwide on 31 August 2026.
- It breaks out AI Overviews, AI Mode, and generative AI in Discover, split by page, country and date, plus device in Search.
- It is free. If you have Search Console, you already have it.
- Catch: impressions only. No clicks, no position. You can see you were shown; you cannot see whether it did anything.
- If your site gets very few AI impressions, the report may show nothing at all. That is not a bug.
So what does that report not tell me?
- Nothing about ChatGPT, Gemini, Claude or Perplexity. It is Google surfaces only.
- No clicks, so no way to calculate a conversion rate from it.
- No prompt text. You learn the page was shown, never the question that pulled it in.
- No competitor view. You cannot see who got named instead of you.
- That gap is exactly the gap the paid tools sell into — which is worth remembering when you read their pitch.
How do I check by hand, for nothing?
- Write down the 15–25 questions a buyer would actually type. Not keywords. Questions, in their words.
- Run each one in ChatGPT, Gemini and Google AI Mode. Log the date, whether you were named, who else was, and whether a link was given.
- Use a logged-out or incognito session, and repeat each prompt twice — answers vary run to run, which is itself the finding.
- Do it monthly on the same list. A trend across three months is worth something; a single snapshot is noise.
- Twenty minutes a month in a spreadsheet. This is the part most people skip and then pay someone else to do.
Do the AI visibility and GEO trackers actually work?
- They work the same way your spreadsheet does: they run prompts and record what came back. There is no privileged data source. Nobody has one.
- So what you are buying is scale and scheduling, not insight — hundreds of prompts a week instead of twenty a month.
- None of them publicly document how they weight or sample sources, which means you cannot audit the number they hand you.
- The honest use is relative: “we were named in 12% of our prompt set in August, 19% in October.” The absolute percentage means very little.
- If your prompt set is small or your category is narrow, the spreadsheet is genuinely as good. Don’t buy the tool yet.
Does llms.txt, AI schema, or content chunking help?
This is where a lot of money is being spent on nothing. Google’s own documentation, updated 10 July 2026, is blunt about it:
- “Optimizing for generative AI search is optimizing for the search experience, and thus still SEO.”
- You “can ignore tactics like chunking content, creating llms.txt files, and seeking inauthentic mentions.”
- And from Google’s AI features page, updated 10 December 2025: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.” No new machine-readable files. No special schema.org markup.
- Plenty of vendors, including agencies in India, are selling an “AI answer-engine kit” built around those exact files. If someone quotes you for llms.txt work, ask them to reconcile it with that page.
- What does carry over: the normal stuff. Crawlable pages, clear titles, a page that answers the question in the first screen, real citations, and genuine third-party mentions.
How much traffic is actually at stake?
Worth having real numbers here rather than the circulating ones.
- Pew Research, published July 2025: across 68,879 searches from 900 US adults, users clicked a traditional search result in 8% of visits where an AI summary appeared, versus 15% where it did not. They clicked a link inside the AI summary in just 1%.
- So AI summaries roughly halve the click-through on affected searches. That is significant and it is not extinction.
- Prevalence depends entirely on how you cut it. SE Ranking found AI Overviews on 82.5% of US branded searches; seoClarity found them on over 62% of queries, ranging 55–84% by industry.
- The headline “AI Overviews appear on over 80% of all Google searches” is a blend of several vendors’ numbers, not a Google figure. Don’t repeat it, and discount anyone who does.
- Assistant traffic is also shifting under your feet. Similarweb’s panel has ChatGPT’s share of generative-AI web traffic falling from ~76% (June 2025) to ~53% (May 2026), with Gemini going from under 9% to roughly 27%. Vendor panel data, so treat as direction rather than fact — but “optimise for ChatGPT” was already the wrong framing.
What should I fix instead?
- Make the answer findable in the first 100 words of the page. Both AI systems and humans reward the page that answers rather than warms up.
- Fix the boring technical layer — indexability, page speed, internal links. There is no AI-specific version of this and Google says so.
- Earn real mentions on sites that actually cover your category. That is what the models were trained on and what they cite.
- Keep your Google Business Profile and category listings accurate if you are local. A lot of assistant answers about Indian businesses are reading those.
- Stop checking weekly. For reference: there has been no Google core update since the May 2026 core update, and the most recent update of any kind was the September 2026 spam update (24 September to roughly 7 October). If your traffic moved outside those windows, it was probably you, not an algorithm.
Is it worth paying someone for this?
| What you’re buying | Real cost | What it actually gets you | The honest limitation |
|---|---|---|---|
| Search Console AI report | ₹0 | Impressions on Google AI surfaces, by page | No clicks, Google only |
| Manual prompt log, 20 prompts/month | ~20 min/month | Named-or-not across ChatGPT, Gemini, AI Mode, with competitors | Small sample, no history until month three |
| AI visibility tracker (SaaS) | Typically billed in USD/month | Hundreds of prompts, scheduled, charted | Undocumented sampling; you cannot audit the number |
| Agency retainer | Quoted per scope | Someone does the work and the fixes | Note that GST on digital marketing services in India is 18% on top; make sure the quote says which |
- Buy the retainer for the fixing, not the measuring. Measuring is nearly free and the fixing is where the months go.
- Ask any agency for the three pages they would change first and why. If the answer is a file format, keep interviewing.
- Expect a slow clock. GrowthLife’s own site says SEO gives “noticeable improvements typically within 3 to 6 months”, which is about right and more honest than most pitches you’ll get.
- No published rate card is normal in the Indian agency market — almost every quote is custom. That cuts both ways, so get two.
What I’d actually do
- Open Search Console today, find the generative AI report, and note your current impressions. That is your baseline and it cost nothing.
- Build the 20-prompt spreadsheet this week and run it. Re-run on the same date each month. Do not buy a tracker until you have three months of your own data to compare it against.
- Ignore anything sold to you as llms.txt, AI schema or content chunking, and put that budget into the page that currently answers your best question badly.