How to Audit How AI Sees Your Brand

What a real audit is for

Most "AI audits" are a handful of screenshots taken on a slow afternoon. Someone asks ChatGPT a few questions, pastes the answers into a deck, and concludes that things are either fine or alarming. It feels productive and it teaches you almost nothing, because a single answer is noisy and a screenshot cannot be re-measured.

A useful audit does two things a screenshot cannot. It covers the prompts your buyers actually use, not the ones that flatter you. And it produces a baseline you can run again, so you can tell whether the work you did moved the answer.

The method below is six steps. You can run a lean version in an afternoon and a thorough one over a week. Either way, the shape is the same.

Step 1: Define the buyer context

Before any prompts, write down who you are auditing for and what they are trying to decide. Skip this and you will end up auditing the questions that are easy to think of, which are almost always branded and almost always misleading.

Pin down four things:

• Personas. The two or three buyer types who actually decide. A founder, a growth lead, a brand lead. They ask different questions.
• Buying moments. Where they are: just discovering the category, comparing options, validating a choice, or handling an objection.
• Constraints. The specifics that change a recommendation: team size, budget tier, industry, region, language.
• The wrong-answer list. Three things AI must not say about you: a competitor you get confused with, a category you do not belong in, a positioning you have moved on from.

That last one matters more than it looks. It turns a vague "is the answer good" into a concrete "did the answer make one of these three mistakes."

Step 2: Build a prompt set that mirrors real buying

The audit is only as good as the prompts. The single most common mistake is auditing branded prompts ("what is [your brand]," "is [your brand] any good"). Those confirm the model has heard of you. They do not tell you whether you appear when a buyer has not already chosen you.

Build a mix, weighted toward the moments closest to a decision:

• Discovery: "best [category] for [persona]," "how do I [job to be done]."
• Comparison: "[competitor] vs alternatives for [use case]," "[your brand] alternatives for [constraint]."
• Use case: "best option for [specific situation]," "what should a [persona] use to [outcome]."
• Trust and objection: "is [category] worth it for [need]," "how reliable is [your brand]."
• Branded: a few, as a control, to see how you are described once the name is known.

Start with twenty for a lean baseline and grow from there. Aim for prompts in the buyer's own words, with their constraints attached. "Most reliable backpack for a rainy weekend hike" beats "waterproof backpack" every time, because it is shaped like a real question.

Step 3: Capture the evidence, not just the verdict

For each prompt, run it across the assistants your buyers actually use, and record more than whether you appeared. The same prompt can return a different answer on a different run, so capture enough to read patterns, not single results.

For every answer, note:

• The full answer, so you can re-read it later in context.
• Your mention: present or absent, and where in the order.
• Competitors named, and which one gets the strongest recommendation.
• The reason the answer gives for its top pick.
• Citations and sources the answer leans on.
• Tone and accuracy: is the description on-strategy, and did it hit anything on your wrong-answer list?

This is the step that separates an audit from a vibe. The reason and the sources are where your fixes come from.

Step 4: Classify what you find

Now turn the captured answers into named gaps. Resist the urge to call everything "we need more content." Different gaps need different teams.

If you have read the other two pieces in this series, this is where they connect. The perception gaps are the visibility vs perception problem in practice, and the competitor losses are the why AI recommends competitors patterns showing up in your own data.

Step 5: Prioritize, do not boil the ocean

A first audit usually surfaces more gaps than any team can fix at once. Prioritization is the whole value, so be ruthless about it.

A simple order that works:

1. High-intent prompts first. A loss on "best [category] for [persona]" matters more than a loss on a broad educational query.
2. Competitor wins before general education. Closing a gap where a competitor is actively recommended over you moves revenue, not just rankings.
3. Wrong facts before new content. Correcting a stale price or category is cheap and high-leverage. Do it before writing anything new.
4. Already-cited pages before net-new pages. Improving a page a model already trusts is faster than earning trust from scratch.

The output of this step is a short, ranked list of fixes with an owner and an expected effect. That is the artifact. Not a deck.

Step 6: Rerun and watch what moves

The audit is not finished when you have a plan. It is finished when you can tell whether the plan worked. Ship the top fixes, wait for the next index cycle, and run the same prompt set again. Compare the new answers to the baseline: did you appear where you were absent, did the description improve, did a competitor's confident line move to you?

This is the part screenshots cannot do. A baseline you can rerun turns AI visibility from a quarterly scare into a loop: measure, fix, re-measure, keep the gains.

The short checklist

If you only remember one thing, remember the shape:

• Personas, buying moments, constraints, and a wrong-answer list, written down.
• A 20+ prompt set weighted toward comparison and use-case, not branded.
• Each answer captured with competitors, reason, sources, and accuracy, across the right assistants.
• Findings classified as visibility, perception, source, citation, or freshness gaps.
• A ranked fix list with owners, high-intent first.
• A scheduled rerun against the same baseline.

From audit to system

Run manually, this is real work, and it is worth doing once by hand so you understand what the numbers mean. The catch is that it does not scale on a spreadsheet. Prompt volatility, competitor tracking, source capture, and reruns add up fast, which is exactly why most teams audit once and never again.

That is the gap Perceptiq closes: it runs buying-moment prompts on a schedule, scores visibility and brand perception, captures the sources behind each answer, and turns the gaps into a prioritized action list you can re-measure. For the model behind the scoring, the AI visibility guide explains how AI decides what to recommend. When you are ready, run your first baseline and start from what the answers actually say.