How do I measure my brand's AI search visibility? The baseline we build for clients
How we measure a brand's AI search visibility for clients: the prompt panel, metrics, source checks, crawler logs and traffic attribution behind our baseline report.
Every program we start begins with a baseline. Before anyone writes a page or pitches a journalist, we measure how the brand shows up in AI answers today. Without that, six months of work cannot be judged. Here is how we build the baseline, and what we use to do it.
Why a screenshot is not a measurement
AI answers vary between runs, between users and between countries. We have watched the same prompt name different brands an hour apart. A single answer is an anecdote. A panel of prompts run daily, averaged over weeks, is a measurement.
Step one: the prompt panel
We write 30 to 50 prompts with the client. Four kinds: category, comparison, problem and brand. We pull wording from sales calls, support tickets and Search Console queries. Promptwatch helps with the last one, since Prompt Explorer suggests prompts from Search Console, ranking keywords, People Also Ask or any page.
Each prompt gets a topic tag and a type. Personas matter when buyers differ, so we stack personas to ask the same question as different people.
Step two: define what counts
Before the first run we set brand aliases and write the Brand Book, whose analysis instructions decide what counts as a mention. A product name, a parent company and a common misspelling all need handling. Changing these later breaks the trend line, so we agree them up front.
Step three: collect and read the numbers
Prompt panels run daily per model, country and language. We report:
- Visibility Score. The headline measure.
- Share of Voice. Against three to five named competitors.
- Position in Answer. Where the brand sits in lists.
- Sentiment. Scored per response, with a weekly AI summary.
Real-interface monitoring covers ChatGPT (Search), Google AI Overviews, Google AI Mode, Copilot, Gemini, Perplexity, Amazon Alexa and Meta AI. Claude, DeepSeek, Mistral and Grok come through APIs, and we say so in the report.
Step four: sources and citations
The numbers say how visible a brand is. Sources say why. Citation analytics list the pages and domains engines cite, with URL positions. Reddit, YouTube and LinkedIn citation reports separate community and video sources. Page Tracker follows specific URLs, ours and competitors'. Citation Trends classify citations by content type and source type, so we know what format engines favor on each topic.
Step five: crawler access
This is the check that surprises people. Agent Analytics shows IP-verified logs of real AI crawlers, including GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot and PerplexityBot. We look for blocked bots, error responses and pages that are fetched but never cited. Crawler logs start on Professional, so they are not part of an Essential baseline.
Step six: traffic and conversions
Visibility that sends nobody is a vanity number. Visitor analytics tracks AI-referred traffic and conversions via a lightweight script, a GTM template or the WordPress plugin. We define one or two conversion events with the client. Volumes are often small. We report them as they are.
Step seven: cross-check with first-party data
We compare against Google Search Console and Bing Webmaster data. Search Insights overlays both, including Microsoft Copilot, on AI crawls and citations, so the cross-check lives in the same place. We do not assume what those sources report about AI features, since it changes, and we check current documentation before quoting anything.
The baseline report
Two pages. Page one: Visibility Score, share of voice, position in answer and sentiment, by topic and engine, with a four-week trend. Page two: top citation gaps, crawler findings, referral counts, and the first five actions. Scheduled PDF reports and Agent Chat help assemble it, and the Action Board carries the follow-up.
What the baseline cannot tell you
It cannot predict that an engine will recommend you, and it does not capture people who ask something you did not put in the panel. We say that to clients before they read the numbers.
Starting plans
The free Explore plan gives a quick ChatGPT-only look with 10 prompts. Essential at $95 a month gives 50 prompts. Professional at $245 adds crawler logs and custom reports, and is where we start most baselines. Agency plans begin at $199 with unlimited projects and prompts.
Other tools do parts of this, and suite add-ons are fine for a light check. For a baseline that includes crawler evidence and traffic, we use the platform that covers all of it.