You can find out whether ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews mention your brand in about 20 minutes, using nothing but a browser. No signup, no trial, no card.
Worth doing before you buy any tracking tool. A manual pass tells you which of the five engines is the actual problem, and that answer changes what you should be spending money on.
What follows is the short version, the exact prompts to use, and an honest note on where the free method runs out.
What can the free check tell you?
Four things: whether you are named, whether you are linked, whether the description is right, and who wins the slots you lose. That covers most of what a first audit needs.
What does it miss?
It cannot give you a trend. One run is one sample on one day, and answers shift when models update. Nothing here proves a direction.
It also cannot scale. Past roughly 30 prompts the manual method collapses under its own tedium, which is the real reason tracking tools exist at all.
How do you run the 20 minute version?
Four steps, in order. Do not skip the clean-session step. It is the one that decides whether your result means anything.
Step 1: Write five prompts
Write five questions a buyer would actually type, with no brand names in the first four. Category questions, not keywords.
The fifth is branded: "What is [your brand]?" That one tests accuracy rather than visibility, and it often surfaces the worst problem in the whole set.
Step 2: Open clean sessions
Use a logged-out or incognito window for every engine. Turn off memory and personalization if the setting exists.
This step matters more than any other. An account that has discussed your company before will name it happily. You conclude you are visible, while a stranger sees nothing.
Step 3: Run and record
Paste each prompt into each engine and copy the full answer into a spreadsheet. Record five columns, not three.
| Column | What to record | Why it matters |
|---|---|---|
| Named | Did your brand appear in the answer text? | This is presence, the headline number |
| Linked | Did the engine cite one of your pages? | This is citation, a separate problem |
| Accurate | Was the description correct? | A wrong description costs more than absence |
| Competitors | Which rival brands appeared? | Your benchmark for a meaningful score |
| Sources | Which domains were cited? | The formats the engine already trusts |
Record the competitor names and cited domains too. Those columns end up being the most useful part of the exercise, because they show which sources the engine already trusts for your question.
Step 4: Score it
Count the answers that named you and divide by the total. Do the same for citations. Keep the two numbers apart.
Presence and citation are different problems with different fixes. Merge them into one score and the whole exercise stops being useful.
Which prompts should you use?
Copy these five and replace the bracketed parts. They are deliberately generic, because generic is what buyers actually type.
- "What are the best [your category] tools?"
- "Who are the leading [your category] providers?"
- "[Competitor A] vs [Competitor B], which should I choose?"
- "I need [the problem you solve]. What are my options?"
- "What is [your brand]?"
Run all five on all five engines. That is 25 answers, and most of the 20 minutes.
Five is enough for a first read. A 20 to 30 prompt set is what you need before the number holds steady enough to track.
What does the competitor column tell you?
The list of domains winning your prompts is the most actionable output of the exercise. Read it before you read your own score.
Which format wins the slot?
Note what shape each cited page takes. Across most categories the same few formats take the slots: numbered checklists, step-by-step procedures with a time estimate, comparison tables with named alternatives.
That repetition is a signal. The engine has already settled on a format it trusts for the question, and a page in a different shape competes uphill no matter how good the writing is.
What if the winner is not a website?
Sometimes the cited sources are video, or a community thread, or a marketplace listing. When that happens on a prompt you care about, no article wins it. Not a better one, not a longer one.
This is more common than it looks. In our July 2026 scan of one audit-related prompt, 6 of the 11 sources Gemini cited were YouTube videos. No amount of written content was going to take those slots.
Those prompts need a different asset entirely. Knowing which ones they are saves a quarter of misdirected writing.
How do you read the result?
The pattern matters more than the score. Three of them cover most outcomes, and each points at a different fix.
What if you appear but are not linked?
Engines know your brand and will not source your pages. A content problem, then, not an awareness one.
The fix is passage-level: direct answers near the top, question-shaped headings, claims that carry a source. That is most of what separates a page that ranks from one that gets cited by ChatGPT.
What if you do not appear at all?
Check access before you write anything. Fetch your own robots.txt and confirm the AI crawlers are not disallowed, then reload a key page with JavaScript disabled. The agent names to look for are published by OpenAI and Google.
If the body copy disappears without JavaScript, retrieval sees the same empty page you just did.
What if the description is wrong?
This is the most urgent result and the least discussed. Wrong pricing, a former product name, the wrong category. Any of those costs you more than absence does.
Wrong descriptions usually trace back to stale third-party pages, old directory listings and press coverage nobody has touched in three years, rather than to anything on your own site. So fixing your homepage does not fix it. The engine was never reading your homepage.
Where does the free method stop?
Three limits, all of which arrive quickly:
- No trend. One run cannot show movement, and movement is the only thing that proves a fix worked.
- No scale. Twenty-five answers is fine; 250 by hand is not.
- No alerting. You will not notice a description going wrong until you next run the check by hand.
The honest read: the free version is diagnostic, the paid version is operational. Run this first, then decide whether what it surfaces is a measurement problem or an optimization problem.
Perplexity and Google AI Overviews diverge most from each other, so a weak result on one says very little about the other.
Where Unveilr fits
The manual check is a snapshot. Unveilr runs the same logic as a loop: scan a fixed prompt set on a schedule, detect which sources won each answer, update the content that lost, then re-scan to confirm the change held.
That last step is the one the free version cannot do. Without a second measurement on the same prompts, a fix and a model update look identical.
In one D2C case study, the brand moved from the 9th most-cited domain in its category to number 1, with ChatGPT visibility rising from 3.3% to 44.7%.