A keyword-level AI recommendation audit takes one afternoon: 15 buyer prompts, four AI assistants, one spreadsheet and a short list of gaps to fix.
Short answer: list 15 high-intent questions your buyers ask, run each one in ChatGPT, Gemini, Perplexity and Claude in a fresh chat, and log whether your brand, a competitor or nobody is named, plus which sources are cited. Every prompt where a competitor appears and you don't is a gap. You need no paid tools for this – the free tiers of the assistants are enough.
What do you need?
A spreadsheet. Google Sheets or Excel.
15 buyer prompts. Not your whole SEO keyword list – the questions someone asks when they're close to choosing a product.
Access to the assistants. ChatGPT, Gemini, Perplexity and Claude. Add Google AI Mode if most of your traffic comes from Google.
Three to four hours. 15 prompts × 4 assistants is 60 answers to read and log.
Step 1: Build your buyer-intent prompt list
The most common mistake is auditing top-of-funnel terms like "what is generative AI". Those rarely lead to a product recommendation. Audit the questions that do:
"best [your category] tools for small businesses"
"[competitor] alternatives"
"[competitor] vs [your brand]"
"how do I [solve the problem your product solves]?"
"[your category] pricing"
Write them as full questions, the way people type into a chat box. "What are good tools for [category]?" gets more recommendation-style answers than a bare keyword.
Step 2: Set up the audit sheet
One row per prompt per assistant, with these columns:
Column | What to record |
|---|---|
Date | Answers change over time, so always date the row. |
Prompt | The exact wording you used. |
Assistant | ChatGPT, Gemini, Perplexity, Claude or AI Mode. |
Brand mentioned? | Yes or no, and the position if it's a list. |
Competitors named | Every brand named, in order. |
Sources cited | The URLs the answer links to, if any. Perplexity, AI Mode and ChatGPT search usually show them. |
Notes | Was anything about you wrong or out of date? Was the tone positive? |
Step 3: Run the prompts in a clean session
Your chat history and saved memories can colour the answers. For each assistant, start a fresh chat with memory or personalization turned off, or use a logged-out session where the assistant allows one. Then, for each prompt:
Paste the prompt exactly as written in your sheet.
Read the whole answer, not just the first list.
Fill in the row, including any sources cited.
Start a new chat before the next prompt so earlier questions don't steer the answer.
For your five most important prompts, run each one twice. If the two answers name different brands, note it: that prompt is unstable, and a single check isn't enough to judge it.
Step 4: Read the results
When the sheet is full, look for patterns:
Which assistant knows you? Count your mentions per assistant. Being named often by Gemini but never by Perplexity points to different source sets.
Which competitor shows up most? Open the sources cited next to their name. Usually it's a comparison page, a review site or a "best tools" list that mentions them and not you. That's what you need to match or get onto.
Which themes are you missing? You might appear for pricing questions but never for "alternatives to" questions.
Is anything wrong? Old prices or retired features are their own kind of gap and often the quickest to fix.
What should you fix first?
Rank the gaps. Start with high-intent prompts where your top competitor is named and you're not. The AI recommendation gap scorecard turns this into a 1–5 score per keyword.
Answer the prompt, don't just target the keyword. Build the page that answers the question completely: who it's for, what it costs, how it compares, with dates on anything that changes.
Re-run the same prompts. Check again two to four weeks after publishing, using the same wording.
To go deeper on why a competitor gets cited and what to publish in response, see how to run a content gap analysis using AI recommendation data. To keep the results in one place over time, use the AI recommendation gap template.
Frequently Asked Questions
How often should I run this audit?
By hand, once a month for your top 15 prompts is realistic. Answers shift between model updates, so a yearly check is too slow. If you need daily or weekly data across more prompts, use a tracking tool.
What if my brand is never mentioned?
That's common for newer or smaller brands, and it gives you a clean baseline. Start with the high-intent prompts where the current answers are weakest – generic advice or no clear winner – and publish a page that answers each one directly.
How is this different from an SEO audit?
An SEO audit checks how your pages rank in search results. This audit checks whether AI assistants name your brand in their answers. They're related – AI answers that use live search draw on pages search engines can find – but you can rank well and still be left out of AI answers.
If you want a quick read before blocking out the afternoon, run the free AI visibility audit. It takes no signup and about 30 seconds, and shows whether AI models know your brand, who they recommend instead and which content gaps to close first. When you're ready to track this every week, blogawesome checks ChatGPT, Gemini, Perplexity and Claude for your keywords and drafts the pages to close the gaps, from $49 a month.