Three prompts to find out whether AI recommends your product, and where you're invisible. Copy, paste, fill in the brackets. Full walkthrough in the newsletter.
If you have any questions or need help, email [email protected] and I’ll guide you through it!
Run this in ChatGPT with web search on. It digs through Reddit, reviews, and the questions people actually search, and hands you back how your customers really talk, in their words.
I'm about to test whether AI recommends my product, and first I need to
know how real customers actually talk about my category. Do real research
and report back in their words, not mine.
- Website: [your URL]
- What it does: [one line]
- Who buys it: [your customer]
Do this:
1. Read my site and pin down my product, price, buyer, main category, and
the narrowest sub-category I could realistically win ("non-toxic
cookware for small kitchens," not just "cookware"). Then list 4-6 real
competitors a buyer would compare me to.
2. Search the web for how real people actually talk: Reddit threads,
review sites (Amazon, Trustpilot, G2), forums, and the questions that
show up in Google's "People Also Ask" and autocomplete for my brand
and category. Pull the EXACT words and worries people use, quote them
verbatim, and cite where each came from. Do NOT invent or polish
anything into marketing language. Only report language you can trace
to a real source.
3. Organize what you find, with verbatim quotes, into:
- how people describe the PROBLEM before they know any brand
- how they describe my product positively
- the main objections and hesitations
- the 2-3 ugliest complaints, word for word
- how they talk about price and value
- the actual questions people search ("is X worth it", "X vs Y",
"alternative to X", etc.)
Keep it tight and skimmable. This is the raw material for my next step.
Paste the report from Step 1 into this. It writes your test set.
Using the customer-language report below, write 15 test questions a real
buyer would type into ChatGPT, Perplexity, or Google.
[paste the report from step 1]
Three groups:
- Problem (5): what someone types before they know my brand exists.
- Comparison (6): "best [category] for [specific type]", "A vs B". Weight
the set here, this is where brands actually get named.
- Objection (4): the last worry before buying, using my two ugliest real
complaints.
For each question give me: the question, its group, whether it contains
my brand name (yes/no), and one word on where the phrasing came from.
Rules: only use phrasing that traces to the report, don't invent it. Skip
plain definition questions like "what is X" (the AI answers those from
memory and never names a brand). Keep them short and conversational, the
way a person actually types.
Finally: tell me the 5 questions I should care about most, and flag any
that are basically unwinnable (an AI answers them without naming any
brand at all).
Take each question through ChatGPT, Perplexity, and Google's AI answer, in an incognito window. Do it by hand, or if you have an AI tool that can browse the web, hand it this:
You're going to check whether a brand shows up when people ask AI what to
buy. Do it carefully.
Brand: [name + website]
Questions: [paste your 15 questions, and mark which ones do NOT contain
the brand name]
Ask each question on ChatGPT, Perplexity, and Google's AI answer. Use a
fresh session each time, and log out / use incognito if you easily can,
so history and personalization don't skew things (if you're already
logged in and logging out is a hassle, just proceed and note it). Run
each question a couple of times, the answers change between runs. Keep
each engine separate, never blend them into one score. Note the date and
which country/currency you're being shown. Before recording an answer,
let it fully load and expand any "show more", picks are often further down.
For each question on each engine, record:
1. Is the brand recommended FIRST, mentioned later, or not at all? (For a
head-to-head or "worth it" question, record the verdict instead of a
rank.)
2. Which competitors got recommended, in the order they appeared.
3. What sources did the answer cite? (Reddit? YouTube? review sites? the
brand's own website?)
4. If the brand is named, is it positive, hedged, or a "don't buy"?
5. Flag anything the answer says about the brand that is factually WRONG
or out of date (wrong price, wrong spec, discontinued product, etc.).
Then give me:
- THE KEY NUMBER: counting ONLY the questions that don't contain the brand
name, how often was the brand recommended first? Report this first, per
engine, and list which questions you counted.
- A leaderboard of which competitors get recommended most.
- A leaderboard of which sources get cited most across all answers. This
is who's currently writing the brand's reputation.
- The biggest gaps (questions where the brand is missing or losing) and
any factual errors worth fixing.
Want more where this came from? Extra Tokens is a weekly newsletter on how to actually get found and grow in the AI era. Subscribe here →