Most brands spend years polishing what they say about themselves. Very few ever check what an AI assistant says about them when a customer asks. Those two things are not the same, and the gap is where trust quietly leaks out.
When a potential customer types your business name into ChatGPT, Perplexity, or Gemini, the model does not read your intentions. It stitches together whatever it can retrieve or recall from third-party sources: directories, mentions, reviews, structured data, and its training memory. The result can be accurate, outdated, incomplete, or simply wrong. And you usually never find out.
An AI detection audit is the disciplined way to find out. Here is how to run one yourself.
Why this matters more than a ranking
Search engine optimization taught everyone to obsess over position. AI assistants change the question. They rarely show ten blue links. They name a few options, describe them, and move on. If the description of your brand is inaccurate, no amount of ranking helps you, because the model is confidently telling people something false.
So the goal of an audit is not "are we recommended?" The honest goal is: is the information about us accurate, current, and consistent across assistants?
Step 1: Write your ground truth first
Before you ask any AI anything, write down the facts you would want a stranger to get right:
- Exact business name and any common misspellings
- What you actually do, in one plain sentence
- Location and service area
- Contact method (email, phone, or booking link)
- Categories you belong to and categories you do not
This is your reference. Without it, you will read AI output and unconsciously grade on a curve.
Step 2: Ask the assistants directly
Run the same set of prompts across ChatGPT, Perplexity, and Gemini. Keep the wording neutral so you are testing detection, not leading the model:
- "What can you tell me about [Brand Name]?"
- "Is [Brand Name] a [category] business? What do they offer?"
- "Who are some [category] providers in [location]?" (Check whether you appear at all.)
- "How do I contact [Brand Name]?"
Run each prompt twice. Models are non-deterministic, and a single answer can mislead you.
Step 3: Score against your ground truth
For every response, mark each fact as one of three things:
- Correct — matches your ground truth
- Missing — the model has no information
- Wrong — the model states something false or outdated
"Wrong" is the most urgent category. A missing brand can be introduced. A misdescribed brand has to be corrected, and correction is slower.
Step 4: Trace where the answer came from
When an assistant cites sources (Perplexity is the most transparent here), follow them. You will usually find the model is echoing a specific directory listing, an old profile, or a page that is not yours. That tells you exactly where to fix the underlying record instead of guessing.
If a model refuses to name a source, treat its claims as unverified memory, not fact.
Step 5: Separate "published" from "detected"
Here is the discipline that most guides skip. Publishing a corrected fact somewhere does not mean the AI now knows it. There is always a lag, and sometimes the model simply never picks it up. So keep two columns in your audit:
- What you have published and verified
- What the AI actually reports back
Only the second column reflects reality for your customers. Re-run the audit on a schedule so you can see whether corrections are actually landing over time.
What an honest audit will and will not do
An audit will show you the current state of your AI visibility and give you a prioritized list of fixes. It will not let you dictate what any model says, and nobody honestly can. Beware anyone promising guaranteed rankings or guaranteed recommendations inside an AI assistant. That is not how these systems work.
The realistic aim is straightforward: make the accurate version of your brand the easiest, most consistent thing for an AI to find, then measure whether detection improves.
Where this fits
This is the core of what we work on at The Resets Company: helping brands become accurately discoverable by AI, even when they do not have their own website. Our approach starts with an AI Detection Audit like the one above, builds consistent presence across credible third-party platforms, and then monitors how AI detection and accuracy change over time. We keep a clear line between what gets published and what AI actually verifies, because that honesty is the whole point.
If you run the audit above and find gaps, that is not failure. It is your starting map. Fix the ground truth, distribute it where it is credible, and measure again.
Top comments (0)