Why one ask isn’t enough
Send an assistant the same prompt twice and it may name different brands. So a single check, by hand or with a free tool, shows you one answer, not how often you’re named.
Each draws on what its model learned in training and, with web search on, on pages it finds while answering. They learn from different data and search differently, so the same brand can be named often by one and rarely by another. Measure each one separately.
Perplexity searches the web for nearly every answer and shows its sources. What it says follows the pages it finds closely, so changes on the web can show up in its answers sooner than in what a model learned in training.
Google’s AI answers sit on top of its search results and draw on them. Sometimes Google shows no AI answer at all; count that apart rather than as a miss, and keep these results separate from the chat assistants’.
Step by step
The same approach works by hand for a first look, or with a tool to keep it going every week.
In their own words, for each kind of person you sell to: “which kids’ walking socks last longest?”, not “kids hiking socks”. Most should not name you. Keep the ones that do, such as comparisons, apart.
Send each prompt to each assistant more than once, because answers vary. Record every answer, including the ones that time out or come back blank.
Work out how often each prompt’s answers name you, then average across prompts. Put a likely range on the result and say how many prompts it rests on.
Check that each match really is you, see who is named instead, and note the sources the answers cite and the prices they quote.
Send the same prompts again later. Call it a change only when the two ranges don’t overlap; otherwise it isn’t established yet.
More prompts help far more than more asks. Asking one prompt nine times still tells you about one prompt, so a rate from 27 different prompts is much firmer than one from 5 asked many times. The guide shows how the likely range narrows as prompts are added.
By hand or with a tool
A first look by hand is worthwhile: a handful of prompts on two or three assistants shows roughly where you stand. Tracking change is another matter. Twenty-seven prompts asked three times on three assistants is 243 answers each time, and they all need reading.
Persipica asks five assistants (ChatGPT, Claude, Gemini, Perplexity and Grok) and Google’s AI answers (AI Overviews and AI Mode) the questions people ask about you, three times each on the chat assistants. It keeps every answer and source, shows how many answers arrived, labels matches it hasn’t verified, and gives every rate a likely range. How Persipica measures sets out the method.
If you’re weighing tools, How Persipica compares puts Persipica next to Profound, Peec AI, Goodie AI and Conductor, with the date each was checked and where each other tool may suit you better. To try a single question first, the free check asks ChatGPT, Claude and Gemini once, with no account.
Next in the library, and two more that go with this one.
Put in your own numbers from your analytics and see what a change in how often AI names you could be worth each month, with the assumptions spelled out.
3 min read GuideWhat GEO is, how ChatGPT, Claude, Gemini, Perplexity, Grok and Google’s AI decide which brands and products to name, and how to measure it honestly.
13 min read ResearchSeven companies were described when asked about by name and not named when a prompt described their category. What that shows, and its limits.
3 min readPersipica asks AI assistants the questions people ask about you, and keeps every answer and source behind each number.