This is the audit our product runs. Below is the manual version, free.

On August 31, Google rolled out generative AI reports in Search Console to every site worldwide. For the first time, you can see how often your pages show up in AI Overviews and AI Mode.

That's a big deal. Google is admitting that "showing up in AI answers" is its own metric now.

But look at what the report leaves out:

• No clicks
• No queries
• No ChatGPT, Perplexity or Gemini
• No competitors

So you know you appeared somewhere, for something, on one engine out of four. You don't know what the AI said, who it recommended instead of you, or which page it trusted.

Here's why that gap matters. On September 21 we ran a SoaRank report for a B2B client: 10 buyer questions, 281 AI queries. They were recommended 0% of the time. Their score was 28/100.

Most people will look at Google's new impressions line, shrug, and move on. Don't. Here's the audit we run to fill in the other three-quarters of the picture. It takes about an hour the first time and about 20 minutes a week after that.

Step 1: Read your new Google report the right way

Open Search Console and go to the generative AI report. Check three things:

1. Which pages earn AI impressions. It's usually not your homepage. It's a guide, a comparison page or a pricing page. Write those URLs down: they're the pages AI already trusts.
2. Which pages earn zero. Your money pages (product, service, pricing) are often missing. That's your first gap.
3. The AI opt-out setting. It now lives in Settings, per property. Make sure nobody on your team switched it on by accident. If it's on, you've removed yourself from AI Overviews and AI Mode.

Step 2: Write 20 buyer prompts, not keywords

Nobody asks ChatGPT "CRM software". They ask:

• "What's the best CRM for a 10-person agency?"
• "Is [your brand] better than [competitor] for small teams?"
• "What do people complain about with [your brand]?"
• "Cheapest alternative to [market leader] that still has [feature]?"

Write 20 of these across four buckets:

1. Category (5): "best [category] for [audience]"
2. Comparison (5): "[you] vs [competitor]", "alternatives to [competitor]"
3. Problem (5): the pain your buyer types before they know your category exists
4. Brand (5): "is [you] legit", "[you] pricing", "[you] reviews"

If you sell locally, add the city to every prompt. "Near me" answers are a different game.

Step 3: Run every prompt on four engines

ChatGPT (with search on), Perplexity, Gemini, and Google (check both AI Overviews and AI Mode).

Use a logged-out or fresh session. Your history changes the answers.

Log each answer in a sheet with five columns:

1. Were you mentioned? (yes/no)
2. Position in the answer (1st, 2nd, buried, absent)
3. Which competitors were named
4. Which URLs were cited
5. Was anything about you wrong? (pricing, features, location)

Run each prompt at least twice. AI answers move between runs, and one run is an anecdote.

Step 4: Calculate your Share of Answer

This is the number that replaces "rank":

Share of Answer = times you were mentioned ÷ total brand mentions across all answers

If four engines named 60 brands in total across your 20 prompts and you were 6 of them, your Share of Answer is 10%.

Now do the same for your top three competitors. That's your real leaderboard, and it almost never matches your Google rankings.

Step 5: Find the citation gap

Go back to column 4 and sort the cited URLs into buckets:

• Your own pages
• Comparison and "best of" lists
• Review sites and directories
• Reddit and forums
• Competitor pages

Then ask one question per prompt: What did the AI read that made it pick them instead of me?

The answer is usually boring. A competitor has a clear comparison page. Or they're on a "best X for Y" list you've never heard of. Or there's a Reddit thread where someone recommended them by name.

That list of URLs is your to-do list. It's more useful than any keyword report.

Step 6: Fix in this order

1. Correct what's wrong. If an engine has your pricing or features wrong, fix the source page first. Put the facts in plain text, not only in images or behind tabs. Wrong answers cost you more than missing ones.
2. Build the page AI is looking for. If competitors win on comparison prompts, publish an honest comparison page: real differences, a table, a clear "who each one is for". AI cites pages that answer the question directly.
3. Answer the questions on your own site. Turn your 20 prompts into a FAQ with short, factual answers. Add structured data (FAQ, Product or Organization schema) so machines can read it without guessing.
4. Get onto the lists that are already cited. If a "best X for Y" article shows up in three engines, reach out to its author. One placement there can beat ten new blog posts.
5. Show up where people actually talk. On Reddit and forums, answer real questions as a real person. No links, no pitch. Engines read those threads.

Step 7: Re-run every week, with the same prompts

This is where most audits die. People fix things and never check whether the answers changed.

Same 20 prompts. Same four engines. Every week.

Log what you shipped next to the scores. After a month, you'll see which fixes moved the answers and which did nothing.

Where this gets hard

To be honest about the limits:

• Answers are noisy. Run prompts more than once before you trust a change.
• You can't buy your way in. No engine sells placement in its organic answers.
• Nobody can promise you the #1 recommendation. Anyone who does is guessing.
• It doesn't scale by hand. 20 prompts × 4 engines × 2 runs is 160 answers a week. That's doable for one brand and painful for five.

That last point is why we built SoaRank. Maya, SoaRank's AI CMO, runs the buyer-question test on ChatGPT, Perplexity and Gemini, gives you a 0–100 score, shows which competitors AI names instead of you, and lays out what to fix first. On the Autopilot plan, she does the work and re-tests every week.

You don't need it to start, though. The sheet above works.

Your move

Google just showed you a quarter of the picture. The other three-quarters is where your buyers are deciding.

If you'd rather not log 160 answers by hand, you can run your first SoaRank report for free at soarank.com.