How do SEO teams measure whether their brand shows up in answers from ChatGPT and Google AI Mode?
How we measure brand visibility in ChatGPT and Google AI Mode for clients: prompt panels, share of voice, citations, crawler logs and traffic, reported as trends.
We have run this measurement long enough to have a standard setup, and the first rule is the same as it was in classic SEO: pick a fixed set of queries and check them on a schedule. Everything else is detail on how to do that when the answer is generated rather than ranked.
Build a prompt panel
We start with the client's real buyer questions, in four groups: category, comparison, problem and brand. We tag them by topic and intent, and by persona and country where it matters. In Promptwatch, the platform we run client programs on, Prompt Volumes and Prompt Explorer help here. Volumes show demand and difficulty, and Explorer suggests prompts from Search Console, ranking keywords, People Also Ask or a page. The panel is typically a few dozen to a couple of hundred prompts, depending on the client.
Measure each engine on its own
ChatGPT and Google AI Mode answer differently, and we do not blend them in the headline number. Promptwatch runs the panel daily per model, country and language, monitoring ChatGPT Search, Google AI Overviews, Google AI Mode, Copilot, Gemini, Perplexity and others through their real UIs. Claude, DeepSeek, Mistral and Grok are tracked through APIs, and we label them that way in reports so nobody assumes they match what a person sees.
Daily runs matter because answers vary. The same prompt can produce a different brand list on a different day. A daily panel turns that noise into a trend.
The metrics we report
Visibility Score, the share of the panel where the brand is named. Share of Voice, against a named competitor set. Position in Answer, which separates first-named from fifth. Sentiment, scored per response from 1 to 100 with a weekly AI summary. Citations, by page and domain. Citation rate per page, from the crawl-to-citation path in Agent Analytics. And AI-referred visits and conversions from visitor analytics.
We set up a Brand Book first, because its analysis instructions decide what counts as a mention. Without it, brands with common-word names produce false positives.
For reporting we use custom dashboards, the A/B period comparison for month over month, and a heatmap that compares any two dimensions, such as competitors against models. PDF reports go out on a schedule, and the Looker Studio connector helps clients who already live there.
Why we add technical and traffic layers
A drop in mentions does not explain itself. We check crawler logs for errors, since IP-verified logs show whether bots like OAI-SearchBot, PerplexityBot, ChatGPT-User or Google-Agent are reaching key pages. We check Search Insights, which overlays Google Search Console and Bing Webmaster data on AI crawls and citations, to see whether classic search moved at the same time. And we look at AI-referred visits to see whether the visibility change reached the business.
Crawler logs start on the Professional plan ($245 a month), so we tell clients on smaller plans what they are missing.
How we talk about accuracy
Panel results are samples, not census data. We explain to clients that a one-day change means little, that engines change behavior without notice, and that we judge trends over four to eight weeks. We do not compare a client's number against an industry average, since definitions of "visibility" differ between tools.
We are also careful about index-style data. Some tools model their prompts from keyword data instead of tracking real prompts, and those numbers can differ sharply from manual checks. For decisions we prefer panels of real prompts, and we spot-check by hand.
What the monthly report contains
Trend lines for the main metrics, a competitor comparison, the biggest movers by topic, the citation sources that changed, the technical issues found, the actions taken and what we plan next. The Action Board in Promptwatch helps with the last two, since it keeps a prioritized list with evidence behind each item.
If you are a small in-house team, you can start the same way on the free Explore plan (10 prompts, ChatGPT only) and move to a paid plan from $95 a month when you need more engines and prompts.