How to Find the AI Prompts Your Competitors Win (and You Don't)

If you want to know which questions AI assistants answer with your competitors' names instead of yours, here's how: write down the prompts your buyers actually ask, run each one many times across ChatGPT, Perplexity, Gemini and Google's AI features, and count who gets named. Every prompt where a rival shows up and you don't is a gap, and every gap is a piece of content waiting to be written. You can do it by hand, with Google Search Console, or with one of the best rated competitor analysis tools for AI search optimization. This playbook covers all three, including the Search Console trap that catches a lot of smart teams.

What an AI search competitor gap analysis actually is

It's the classic content gap analysis SEO teams already run, pointed at a new surface. The old version asks, "Which keywords do competitors rank for that we don't?" The new version asks, "Which buyer prompts get answered with a competitor's name and not ours?"

Three things change when you make the switch:

• The unit is a prompt, not a keyword. Buyers type full questions with context baked in: budget, team size, industry, the tool they're trying to leave.
• The score is frequency, not rank. AI answers reshuffle constantly, so "position 3" means very little. "Named in 6 of 10 answers" means a lot.
• There are two ways to win. A mention is your brand name appearing in the answer. A citation is the AI using your page as a source. You want both, and you can be missing either one. Our guide to AI search visibility metrics breaks down the difference.

The output hasn't changed from any content gap SEO project: a ranked list of pages to publish.

Step 1: Map the prompts your buyers actually ask

Start with a prompt set that covers every stage of the buying conversation, with several prompts per cluster. Too few and you'll miss whole categories of demand. Too many and you'll never finish the manual version.

We group prompts into six clusters in our own AI visibility dashboard, and they work nicely as a starting grid for anyone:

Where do the real prompts come from? Sales call recordings, support tickets, demo request forms, and the "how did you hear about us" field are all gold. So is one free source you already have open in another tab.

Step 2: Mine Google Search Console for real prompts

Search Console won't show you what people typed into ChatGPT. It does show you something close: the long, conversational queries people now type into Google, including Google's own AI features. Google's documentation confirms that impressions and clicks from AI Overviews and AI Mode are counted in the Performance report, and that a follow-up question inside AI Mode is logged as a brand new query.

Here's the method:

1. Open Performance, then Search results, and set the date range to the last three months (skip comparisons that reach back before late April 2026, for reasons covered below).
2. Add a Query filter and choose Custom (regex).
3. Paste ^(?:\S+\s+){9,}\S+$ to show only queries of ten words or more. This is the pattern Search Engine Land published for surfacing prompt-style searches.
4. For a second pass, try a question-shaped filter such as ^(who|what|when|where|why|how|which|can|does|is|are|should|best|compare)\b(\s\w+){5,}, which catches queries that open with a question word and run at least six words long.
5. Export the lot and sort it into the six clusters from Step 1.

One thing to know before you go hunting for a shortcut: Google rolled out a dedicated Generative AI performance report in 2026 that isolates impressions from AI Overviews and AI Mode. It's useful for seeing which of your pages surface in AI answers, but as Search Engine Roundtable noted, it has no query data and no click data. It also only appears for properties with enough data, so you may not see it at all. The regex trick is still how you get to the actual wording.

Also remember that length is a hint, not proof. Plenty of long queries are ordinary Google searches, and plenty of valuable prompts are short. Treat this export as a list of candidate prompts to test in Step 3, not as a finished map.

The Search Console catch: some of those "prompts" are bots

This is the part most walkthroughs skip, and it matters. Not every long, chatty query in your export was typed by a human with a credit card. Some of it is automated: rank trackers and other tools scraping Google, plus prompt-shaped queries that read like machine-written briefs.

We know machines can move Search Console numbers because we've all watched it happen. When Google switched off the &num=100 results parameter in September 2025, a favorite of scraping tools, an analysis of 319 properties found that 87.7% of sites lost impressions and 77.6% lost unique ranking terms. Clicks didn't fall with them. Those vanished impressions were never people.

Prompt-shaped queries have the same problem. One site that published its own Search Console breakdown found that 83 of the 265 queries Google disclosed over 28 days were prompt-shaped, including role-prefixed paragraphs and 77-word vendor briefs, with an average position around 72 and zero clicks. That's a single property and not a benchmark, but it's a pattern worth knowing on sight.

So before a query earns a spot on your prompt map, run it through a quick sniff test:

• Clicks beat impressions. A long query with clicks is far more likely to be a real person. Sort by clicks first and start there.
• Zero clicks at a deep position is a yellow flag. Before num=100 went away, tools fetching 100 results at a time earned impressions for pages buried far down the results, and that one site's prompt-shaped queries averaged position 72 with no clicks.
• Read the phrasing. "You are a procurement analyst…", pasted URLs, numbered instructions and paragraph-length briefs are how people talk to machines, or how machines talk to each other. They are not how people talk to Google.
• Look for families, not singletons. Real demand shows up as several human-sounding variants of the same question. A lone, oddly precise, paragraph-length query with a single impression deserves suspicion before it earns a spot on your map.
• Don't trust impression trends across the logging bug. Google's data anomalies page confirms a logging error misreported impressions from May 13, 2025 to April 27, 2026. Clicks were unaffected, which is one more reason to lean on them.

Bot-flavored queries aren't worthless, by the way. A machine-written prompt about your category still tells you which topics are being checked. Just don't size your content plan on its impression count. Use Search Console to collect wording, then let repeated testing in the AI engines tell you where the gaps are.

Step 3: Run each prompt repeatedly and count who gets named

Now the fun bit. Take your prompt list to ChatGPT, Perplexity, Gemini, Claude and Google's AI Overviews, and record which brands each answer recommends.

The word that matters is repeatedly. In research from SparkToro and Gumshoe covering 2,961 runs of 12 prompts, ChatGPT and Google's AI returned the same list of brands less than 1% of the time when asked the same question again. The same list in the same order showed up less than 0.1% of the time. One run tells you almost nothing.

What does hold steady is how often a brand appears. In the same study, the leading headphone brands showed up in 55% to 77% of responses. That's your metric. For each prompt, log:

• How many runs out of the total named you
• How many named each competitor
• Whether the description of you was accurate
• Which URLs the engine cited

Then calculate share of voice for each cluster: your mentions divided by total mentions for all brands. A prompt where a competitor appears in most runs and you appear in none goes straight to the top of the list.

By hand you'll manage a handful of runs per prompt before patience runs out. That can hint at a gap, but it isn't statistically robust. The study ran each prompt 60 to 100 times per platform to see stable patterns, which is where software earns its keep.

Step 4: Trace the sources behind each competitor win

When a competitor wins a prompt, find out why. Look at what the engine cited and check for one of three situations:

1. Their own page answers the question and yours doesn't exist. This is a content gap, and it's the easiest kind to close.
2. A third party vouches for them. Think review sites, roundups, community threads and editorial coverage. Profound's analysis of 680 million citations found Wikipedia was ChatGPT's most-cited source and Reddit led for Perplexity, though exact shares vary a lot from study to study. This is an authority gap.
3. You have a page, but it isn't getting picked. Usually the answer is buried, vague, or missing the specifics the prompt asked for. This is a structure gap.

There's a fourth, sneakier cause. Google says AI Overviews and AI Mode may use "query fan-out", issuing multiple related searches across subtopics to build one answer. That rewards depth: in Surfer's study of 10,000 keywords, pages that ranked for the main query and its fan-out queries were 161% more likely to be cited in AI Overviews than pages ranking for the main query alone. So when a rival wins a broad prompt, check whether they also cover the subtopics around it.

Step 5: Turn the gaps into a content plan

Score every gap on three things: buyer intent (is this person shopping or browsing?), how firmly competitors hold it, and whether you have a credible claim to the answer. High intent, wobbly competition and real expertise is your first sprint.

Then match each gap to the format that fits it:

Specific beats broad every time. When we worked with Runway, the content was built around the exact questions finance teams ask, and Runway now appears in 92% of AI answers for "What's the best modern FP&A platform for fast-growing startups?" The Barry Law Firm got pages for RV claims, fleet vehicles and individual car brands, and won 40 AI answers it had never appeared in before within two months. Neither result came from one giant guide. They came from many precise answers to precise prompts.

Finally, re-run your prompt set every few weeks. Competitors publish, models update, and yesterday's win can quietly wander off.

The best rated competitor analysis tools for AI search optimization, compared

Running 50 prompts across five engines dozens of times each gets old fast, so it's worth automating Step 3. Here's what the widely used trackers offer, based on their published pricing pages at the time of writing. Prices change often, so check before you buy.

So which is the best platform for AI search optimization competitor analysis? It depends on what happens after the dashboard loads. Every tool above will show you where you're missing. If you have writers and a strategist ready to act on the list, a tracker on its own is a smart buy. If the list is going to sit in a tab while everyone's busy, you need the finding and the fixing in one place. We compare more options in our roundup of tools that show how your brand appears in AI search.

How we run this at Hatter

We built Hatter because we kept doing this analysis by hand as an agency, and the gap between "found it" and "fixed it" was where results went to nap.

Our AI visibility monitoring runs representative prompt sets repeatedly across ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews, and flags when your brand gets named, ignored, or replaced by a competitor. You see share of voice against the direct competitors that matter in your market, which source URLs each engine pulled in, and how it all shifts over time. It's sampled buyer prompts through repeated runs, not a recording of every conversation anyone has ever had with an AI, and we think that honesty is the point: frequency across many runs is the number you can trust.

Then the agents get to work. Our AEO Agent mines the prompts your customers ask AI, spots where competitors are being recommended, and drafts the content that closes the gap. Our SEO Agent does the same with live keyword data for Google. Each opportunity arrives with the prompt or keyword, the buyer intent, the competitors currently winning it, and a likely-impact score. Drafts are written in your brand voice, wait for your approval, and publish to a Hub on your own domain that captures leads, with no developer and no CMS migration. Because the agents run continuously, new gaps show up as the market moves, not at the next monthly meeting.

If you'd like a strategist steering the whole thing, Hatter Managed adds deeper strategy, execution and reporting, with human review of every piece. Both options are on our pricing page.

Curious where you stand right now? Our free SEO and AEO audit shows how you compare with competitors in AI answers and surfaces your first prompt and keyword opportunities. No card required.

Frequently Asked Questions

How can I find which questions AI recommends my competitors for instead of me?

Build a list of buyer prompts across best-of, alternatives, comparison, pricing, use-case and implementation questions. Run each one several times in ChatGPT, Perplexity, Gemini and Google's AI features, and record how often each brand is named. Prompts where a competitor appears in most runs and you appear in few or none are your gaps. An AI visibility tool automates the runs and tracks the results over time.

Can Google Search Console show me AI prompts?

Partly. Search Console shows the queries people type into Google, including AI Mode and AI Overviews, but not what they ask ChatGPT or Perplexity. Filtering the Performance report with a regex for queries of ten or more words surfaces prompt-style searches you can add to your testing list. Google's Generative AI performance report shows impressions by page but doesn't include queries.

How do I tell bot queries from real ones in Search Console?

Lean on clicks, since real people click and scripts mostly don't. Be suspicious of long queries with zero clicks at deep positions, queries that read like instructions to a chatbot, and queries containing pasted URLs or role prefixes. Several human-sounding variants of the same question signal real demand. A single hyper-specific paragraph with one impression deserves a second look.

How many times should I run each prompt?

More than once, always. AI answers vary so much between runs that a single response isn't reliable. A handful of manual runs can hint at a gap, but it isn't robust. The SparkToro and Gumshoe research ran prompts 60 to 100 times per platform to see stable patterns, which is realistic only with software.

How is this different from a traditional content gap analysis?

A traditional analysis compares keyword rankings on Google. An AI search gap analysis compares how often brands are named and cited in generated answers to full-sentence prompts. The two overlap heavily in practice, because clear, specific pages that answer buyer questions tend to help on both fronts, which is why we run SEO and AEO from the same content engine.