Why does my firm show up in ChatGPT but not in Google's AI answers?

That is closer to literal than it sounds.

Kevin Indig’s H1 2026 analysis found that roughly 91% of citations appear in only one engine. A separate study across 161,000 prompts found the major engines share only about 17% of their cited sources, with around 70% of cited sources appearing on a single engine.

So being named by one engine tells you very little about your standing in any other. These systems are not converging, and the gap is not a bug you can tune your way out of. It is the current state of how they retrieve.

What a single score costs you

Imagine a firm cited constantly by Perplexity, occasionally by Google’s AI Mode, and never by ChatGPT or Gemini. Averaged, that might read as moderate visibility, which sounds like a tuning problem.

It is not a tuning problem. It is a total absence from two engines and a strong position in one, which calls for completely different work than “improve the score.” The average has destroyed the only actionable information in the data.

Presence is a yes or no question, per engine, before it is ever a rate question.

The audience is moving too

The distribution is not stable, which makes single-engine strategies riskier than they look. ChatGPT’s share of AI search activity fell from roughly 78% to 56% in six months. Gemini rose to around 30%. Claude sits near 10%.

A firm that optimized entirely around ChatGPT eighteen months ago was optimizing for a shrinking slice and would not have seen it in a blended number.

How to measure it properly

Build a real prompt set. Not keywords. The questions a hurt person actually types, in their words. “Do I have a case if I was rear-ended and felt fine at the scene.” “How much does a personal injury lawyer cost in Sacramento.” “Should I accept the insurance company’s first offer.” Thirty to fifty of them, covering the situations your firm actually takes.

Run them per engine, separately. ChatGPT, Google AI Overviews, Google AI Mode, Perplexity, Gemini, Claude. Record three things per prompt: were you mentioned, were you cited with a link, and which sources the engine used instead.

That third column is the valuable one. It is a map of the evidence layer for your market, and it usually explains the other two columns.

Track it as a grid, not a number. Rows are prompts, columns are engines. What you want to see is where the blanks cluster. Blanks concentrated in one engine is a source-pool problem. Blanks spread evenly across a topic is a content gap. Those need different responses, and a single score cannot tell them apart.

Add Search Console where you have it. Google’s Generative AI performance report, impressions only, broken out by page, country, and device, is currently the only first-party data Google provides on AI Overviews and AI Mode performance. Rollout is site by site, so check rather than assume. If it is there, start baselining now so there is a before to compare against.

What this costs

More than a dashboard subscription and less than most firms expect. The prompt set takes an afternoon to build properly and gets reused. The run is repeatable and largely automatable. The judgment, reading the source column and deciding what it means, is the part that takes actual expertise and does not automate.

Monthly is the right cadence for most firms. Weekly is noise at this sample size.

The point

You are not trying to win “AI.” There is no AI. There are five or six systems with different source pools, different refresh behavior, and different audiences, and your firm has a different standing in each one.

Measure them separately or you will keep optimizing against an average that describes none of them.