Aizii Learn · AI Search

[← Back to Learn](https://aizii.ai/learn)

# How to find out what AI models say about your brand

Ask the models directly, with the questions your buyers ask as well as questions that name your brand, and run the same questions across ChatGPT, Gemini, Perplexity, and Claude. Log what each one says, including which sources it cites. No single dashboard shows all of it. Google Search Console reports on Google's AI features only, so the rest of the picture comes from asking and recording.

## Why one check is not enough

Each assistant draws on different sources, so being named in one says nothing about the others. Gemini is grounded on Google's index. ChatGPT search reads live pages through OpenAI's own search. Perplexity searches the live web and shows its sources. Claude answers from its own retrieval. The same error can exist in one assistant and not in another, and the same brand can be recommended by one and invisible to another.

Answers also vary from run to run. Ask the same question twice and the wording, the order, and sometimes the shortlist can change. A single check is a snapshot, not a measurement. The value is in the comparison across runs.

There is a second split to watch. An answer that carries citations came from pages the assistant retrieved. An answer with no citations may come from the model's training memory. Those are two different failure modes with two different fixes, so note which kind you are looking at before you try to fix anything.

## Step 1: Build a fixed prompt set

Write the prompt set once, then do not change it between runs. Changing prompts destroys comparability. Three kinds of prompt cover the ground:

1.  Brand-name prompts. What does the model know about you directly. Examples: "What is \[your brand\]?" "What does \[your brand\] sell, and who is it for?" "Is \[your brand\] any good? What are its pros and cons?"
2.  Buyer prompts that never name your brand. This is the step most brands skip. Buyers describe a problem, not a brand. Examples: "What are the best \[category\] products for \[your buyer\]?" "What are the alternatives to \[your biggest competitor\]?" "\[Competitor A\] vs \[competitor B\], which should a small team pick?"
3.  Fact prompts. These surface wrong details. Examples: "How much does \[your brand\] charge for \[product\]?" "Where is \[your brand\] located?" "What is \[your brand\]'s return policy?"

Ten to twenty prompts is a workable range. Cover your main category, your main buyer, and your main competitor by name.

## Step 2: Run the set clean, on each assistant separately

Run each prompt in a fresh chat, so an earlier conversation about your brand does not shape the answer. Run the full set on one assistant before moving to the next, and keep search on where the assistant offers it.

Note two things as you go. First, whether the answer used search. If it includes citations, the web was used. If it does not, you may be looking at the model's memory. Second, whether your brand was named at all, and in what position. A brand named last with no reason given is a different result from a brand named first with a specific strength attached.

## Step 3: Log five things for every prompt

For each prompt and each assistant, record:

-   Named or not, and in what position
-   What was claimed about your brand, in the model's own words
-   Whether each claim is accurate, outdated, or wrong
-   Which sources were cited, if any
-   Which assistant gave the answer

Keep the log in one sheet so runs can be compared side by side. The log is the audit. Without it, you have impressions, not evidence.

## Step 4: Trace each error to its source

Sort errors into two piles before fixing anything.

A retrieval error happens when the assistant searched, found a page containing wrong information, and repeated it. Open the cited page and the wrong claim is usually sitting there in plain text. Fix the source page, and the answer has a chance to change on the next retrieval. If the cited page is a review site, a roundup, or a forum you do not control, the fix is a correction or a listing there, or a stronger page of your own that the assistant can retrieve instead.

A training error happens when the model states something from its own memory, with no source behind it. There is no direct edit path into a model's memory. The fix is indirect: make the correct facts consistent and easy to retrieve everywhere the model reads, on your own pages, in your listings, and in structured data, then re-check on a later run. Do not expect a training error to change the same week you fix a page.

## Step 5: Repeat the same set on a regular schedule

Run the same prompts, on the same assistants, on a regular schedule, and compare each run against the last. Watch for three movements: your brand appearing where it did not before, wrong claims disappearing, and competitors appearing where you used to appear. A brand that appears once proves it can happen, not that it usually does. Frequency across runs is the measure that matters.

### The fundamentals

These mechanics are shared across every engine and covered in full in their own guides:

-   •The llms.txt format and a complete sample, see our [guide to writing a good llms.txt file](https://aizii.ai/learn/ai-search/how-to-write-a-good-llms-txt-file)
-   •The markup assistants read to verify facts, see our [product schema guide](https://aizii.ai/learn/ai-search/how-to-add-product-schema-markup-for-ai-search)
-   •The complete crawler rule reference, see our [robots.txt guide for AI search](https://aizii.ai/learn/ai-search/how-to-write-robots-txt-for-ai-search)

### Say you run a skincare brand

Say you run a skincare brand that sells fragrance-free moisturizer for sensitive skin. You run your prompt set and find three different problems. ChatGPT recommends a competitor and never names you on buyer prompts. Perplexity names you but cites a review roundup that leaves you out of its top picks. Gemini repeats an old price from a page you updated months ago. Each problem has a different source: a missing presence in the places ChatGPT retrieves, a roundup that needs a listing or a correction, and an outdated page still ranking in Google's index. One prompt set, three fixes, none of them visible from a single check.

## What this guide does not cover

This guide covers finding out what the models say. It does not cover rewriting the pages the models retrieve, getting listed on the review sites they cite, or the transaction layer that turns a recommendation into a sale. It also does not cover paid tracking tools, which run this same method on a schedule for you.

## Check your own site now

Run the free scan. One click, no signup, and it shows where your site stands with every AI crawler today: aizii.ai/aeo-scan

[Run the free scan](https://aizii.ai/aeo-audit)

## Sources

-   Google AI features and your website (specific): [https://developers.google.com/search/docs/appearance/ai-features](https://developers.google.com/search/docs/appearance/ai-features)
-   Google guide to optimizing for generative AI features (specific): [https://developers.google.com/search/docs/fundamentals/ai-optimization-guide](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide)
-   OpenAI bots documentation (specific): [https://platform.openai.com/docs/bots](https://platform.openai.com/docs/bots)
-   Perplexity bots documentation (specific): [https://docs.perplexity.ai/guides/bots](https://docs.perplexity.ai/guides/bots)

Note: links marked specific point to the page that supports the claim in this guide. Links marked general docs point to the vendor's general docs section for that topic, not to a page that proves every step. Check the vendor docs for the current path before relying on a menu name or click path in this guide.
