You ask ChatGPT for the best tools in your category and get a confident list of four. Your two biggest competitors are on it. You aren't. You try Google's AI Mode and Perplexity: same story, give or take. It's tempting to write this off as a black box. It usually isn't. AI answers are assembled from sources you can see, and the reasons a rival gets named are often sitting in the citations.
A competitor gap analysis for AI search is the structured version of that detective work. You choose prompts that reflect how buyers compare options, record who appears and how they're described, trace the sources behind each mention, classify the gaps, and build a plan to close the ones that matter. Then you re-test, carefully, because AI answers vary from run to run.
This guide walks through each stage with templates you can copy into a spreadsheet. It assumes you already know the basics of tracking AI visibility; if not, start with our guide to tracking your brand's visibility in AI search.
Building a prompt set that reflects how buyers compare you
The prompt set is the foundation, and most gap analyses go wrong here. Teams test the prompts they would like to win ("best [category] software") and ignore the prompts buyers actually use. Build the set from real language: sales call notes, support tickets, on-site search logs, People Also Ask boxes and forum threads.
| Prompt category | Example | Share of the set | What it reveals |
|---|---|---|---|
| Category shortlist | "Best project management tools for agencies" | 25–30% | Whether you're in the default consideration set |
| Constraint-led | "Project tool with client portals under $20 per user" | 25–30% | Whether your differentiators are understood |
| Head-to-head | "[Competitor] vs [you]"; "Is [you] better than [competitor] for agencies?" | 15–20% | How you're described relative to rivals |
| Alternatives | "Alternatives to [competitor]" | 10–15% | Whether you're seen as a substitute |
| Problem-first | "How do I stop agency projects going over budget?" | 10–15% | Whether your content is used to explain the problem |
Aim for 30 to 60 prompts per product line. Fewer than 20 and a single odd answer distorts the results. Write each prompt the way a person would type or speak it, include the locations or segments you serve, and freeze the set: you can't measure progress if the questions keep changing.
Choose your competitors deliberately too. Include two or three direct rivals, plus any brand that keeps appearing in answers even if you don't consider it a competitor. Assistants often surface review sites, marketplaces or adjacent tools as answers, and those are competitors for attention.
Recording mentions, position and description in a scoring sheet
Run each prompt in each assistant you care about (typically ChatGPT, Google AI Mode or AI Overviews, Perplexity, and Gemini or Copilot), in a fresh session without personalisation. Run each prompt at least twice, because answers vary. Record the results in a sheet with one row per prompt, per assistant, per run.
| Column | What to record | Scoring |
|---|---|---|
| Mentioned? | Is each brand named in the answer? | 1 = yes, 0 = no |
| Position | Order of mention in a list | 1st = 3 points, 2nd–3rd = 2, later = 1 |
| Recommended? | Is the brand recommended for the user's situation, or only listed? | Recommended = 2, listed = 1 |
| Description | The exact words used to describe each brand | Paste verbatim |
| Accuracy | Are features, prices and facts correct? | Correct / partly / wrong |
| Cited sources | Every URL the answer cites or lists under sources | Paste URLs |
| Links work? | Do links to your site resolve? | Yes / 404 |
From this you can calculate a simple share of voice for each brand: mentions divided by total answers, and a weighted score using position and recommendation points. The description column is where qualitative gaps show up. If every answer calls a competitor "the best option for agencies" and calls you "a simple tool for freelancers", you have a positioning problem that no amount of extra mentions will fix.
The "links work?" column catches a separate issue: assistants sometimes cite URLs on your site that don't exist. If you see this often, our set guide to fixing AI hallucinated URLs covers it.
Tracing the sources behind each competitor mention
This is the step that turns a scoreboard into an action plan. When an assistant searches the web, it usually shows its sources. OpenAI's help centre describes a Sources button in ChatGPT that lists cited and related links, and Perplexity and Google's AI features show citations alongside answers.
- Collect every cited URL from answers where a competitor appears and you don't.
- Classify each URL by type: competitor's own site, third-party list or review article, review platform, forum or community thread, video, news or press, documentation or directory.
- Count by domain. If one comparison article appears in 40% of answers, it's a key source for your category.
- Open each high-frequency source and check whether you're mentioned in it, and how. Often you're simply missing from the list the assistant is summarising.
- For answers with no citations (common when an assistant answers from its own knowledge), search the competitor's description phrase in quotes. The wording often traces back to a specific page that has been widely repeated.
- Note what the competitor's own cited pages have that yours don't: a dedicated page for the use case, a pricing table, a comparison page, a recent date.
This is the same logic as co-citation in classic SEO: brands that are repeatedly mentioned together in trusted sources get associated with each other and with the topic. Our set guides on co-citation and co-occurrence and surround sound SEO go deeper on earning presence across the pages that rank for your category.
Five gap types and how to tell them apart
Once the sources are classified, most gaps fall into one of five types. Each has a different owner and a different fix, so naming the type early saves wasted effort.
| Gap type | Evidence in your data | Typical fix | Usual owner |
|---|---|---|---|
| Access gap | Your pages never appear as citations, even for branded prompts | Check robots.txt and CDN rules for search crawlers such as OAI-SearchBot; fix indexing issues | Technical SEO / dev |
| Content gap | Competitor's own pages are cited for use cases or constraints you also serve, but you have no matching page | Create focused pages for those use cases, comparisons and pricing | Content |
| Third-party gap | Cited lists, reviews and articles mention competitors but not you | Digital PR, outreach to list authors, review site profiles | PR / partnerships |
| Reputation gap | You're mentioned but with caveats ("some users report…") drawn from reviews or forums | Fix the underlying issue, respond publicly, generate fresh reviews | Product / customer success |
| Positioning gap | You're mentioned but described for the wrong segment or with outdated facts | Update your site, profiles and key third-party pages with consistent, current positioning | Marketing / brand |
The access gap is quick to rule out and occasionally explains everything. OpenAI's crawler documentation is explicit:
“Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers.”
— OpenAI, Overview of OpenAI Crawlers
If a security rule is blocking it, no amount of content work will help in ChatGPT.
The third-party gap is the most common in competitive B2B categories. Our guides on getting into "best of" lists and digital PR for AI visibility cover the outreach side.
Why the gap looks different in Google's AI features and in ChatGPT
Don't be surprised if you're competitive in one assistant and absent from another. The systems retrieve from different places and appear to weight brand signals differently.
Ahrefs' July 2025 analysis of the top 50 brands in its Brand Radar data found that branded web mentions correlated strongly with visibility in Google AI Overviews (ρ = 0.65) but only weakly in Perplexity (0.30) and very weakly in ChatGPT (0.15). Its larger study of 75,000 brands, published in December 2025, found branded web mentions among the strongest correlates across ChatGPT (0.664), AI Mode (0.709) and AI Overviews (0.656), with YouTube mentions correlating at about 0.737.
“The usual disclaimer applies: correlation isn't causation.”
— Ahrefs, Top Brand Visibility Factors in ChatGPT, AI Mode, and AI Overviews
The practical lessons for a gap analysis:
- Segment results by assistant. A blended share of voice hides the fact that you may have a Google problem and a ChatGPT problem with different causes.
- Treat Google's AI features as an extension of Google Search. Google says AI Overviews and AI Mode use its core ranking and quality systems, so organic rankings for the underlying queries matter.
- Don't over-index on mentions alone. The big-brand advantage in Google's AI features is partly a reflection of overall brand size. Smaller brands often find more room in prompts with specific constraints.
- Check YouTube. If competitors have reviews and tutorials on YouTube and you don't, that's a gap worth recording even if you can't prove its effect.
Worked example: a project tool versus two larger rivals
Hypothetical example; brand names and figures are invented to show the method.
"Planlight" sells project management software for creative agencies. It runs 40 prompts across four assistants, twice each: 320 answers in total.
| Brand | Mentioned in | Share of voice | Typical description |
|---|---|---|---|
| Rival A | 214 answers | 67% | "Best all-round option for agencies" |
| Rival B | 173 answers | 54% | "Strong for client approvals and proofing" |
| Planlight | 58 answers | 18% | "Lightweight tool for freelancers" |
Source tracing finds three things. First, two comparison articles appear in 38% of answers that name Rival A; neither mentions Planlight. Second, Rival B is cited for "client approvals" from its own feature page; Planlight has the same feature but describes it only in its help centre. Third, the "freelancers" description traces to Planlight's own homepage from two years ago, copied into several directory listings.
That gives three gap types: third-party (the comparison articles), content (no client-approvals page) and positioning (outdated freelancer messaging). The access check is clean.
Turning findings into a 90-day action plan
Prioritise gaps by how many answers they affect and how hard they are to fix. A spreadsheet with "answers affected", "effort (1–5)" and "owner" columns is enough.
- Days 1–14: fix positioning at the source. Update the homepage, About page, product pages and every directory or review profile you control with consistent, current positioning. This is cheap and influences every assistant.
- Days 1–30: close content gaps. Publish focused pages for the use cases and constraints where competitors' own pages are cited. Include specifics: features, pricing, integrations, who it's for. An honest comparison or alternatives page often earns citations for head-to-head prompts.
- Days 15–90: work the third-party sources. Contact authors of the most-cited lists with a useful update (new features, data, a trial account). Pitch original data or expert commentary to publications in the category.
- Days 30–90: address reputation themes. If forums or reviews repeat a complaint, fix the issue and say so publicly, then ask recent customers for reviews.
- Ongoing: build presence on platforms competitors own. YouTube reviews, community threads, podcasts. Slow, but they compound.
Our guide to content gap analysis covers finding missing topics in classic search, and the set guides on AI visibility for new brands and AI search for mobile apps adapt this plan for those situations.
Re-testing without fooling yourself
AI answers are probabilistic. The same prompt can name different brands on consecutive runs, and assistants update their models and retrieval often. To tell real progress from noise:
- Keep the prompt set frozen and run it on a fixed schedule, ideally monthly.
- Run each prompt at least twice per assistant and use the average.
- Look for movement of more than about 5 percentage points in share of voice on a set of 30+ prompts before celebrating. Smaller moves are often noise.
- Track the description column as closely as the mention count. A shift from "for freelancers" to "for agencies" is real progress even if share of voice moves slowly.
- Log changes (page launches, PR placements, model updates you're aware of) on the same timeline so you can connect causes and effects.
- Connect to traffic. GA4 added an AI Assistant channel in May 2026, but Search Engine Journal noted that some assistants, including Perplexity, still land in Referral, so filter by source too.
Tools can automate the runs and the scoring, but a hand-run analysis of 40 prompts is a good first pass, and it forces you to read the answers. Our guide to using AI for competitor analysis covers some of the tooling.
Questions about AI competitor gap analysis
How many prompts do I need?
For a single product line or service, 30 to 60 prompts across three or four assistants usually gives a stable picture. Fewer than 20 and one unusual answer can swing your share of voice by several points.
Why does a competitor appear in ChatGPT but not in AI Overviews, or the reverse?
The systems draw on different retrieval sources and weight signals differently. Ahrefs' 2025 data showed branded mentions correlating far more strongly with Google AI Overviews visibility than with ChatGPT visibility.
Can I see exactly why an assistant recommended a competitor?
Not exactly. Cited sources and the wording used are strong clues, not proof.
Should I copy the content that gets competitors cited?
No. Identify the question it answers and the evidence it provides, then answer the same question better from your own experience. Google's guidance favours original, non-commodity content.
How long before closing a gap shows up?
Site fixes can appear within weeks of recrawling. Third-party coverage and reviews usually take a quarter or more.
Running your first AI gap analysis this month
Write 40 prompts from real customer language, run them twice in four assistants, and fill in the scoring sheet. Spend most of your time on the sources column: the two or three pages that appear again and again will tell you more than any dashboard. Then pick one fix from each gap type and put an owner and a date against it.
If you'd like an outside team to run the analysis and the follow-up work, our generative engine optimization guide explains our approach, and you can get a quote for an AI visibility audit.
Further reading on AI search competitor gap analysis
- AI Visibility for New Brands: Getting Mentioned With No Track Record
- AI Search for Mobile Apps: Get Recommended When Users Ask for an App
- AI Hallucinated URLs: Fix 404s From ChatGPT and AI Referrals
Sources for this AI competitor gap analysis guide
- Ahrefs: Top Brand Visibility Factors in ChatGPT, AI Mode, and AI Overviews (December 2025)
- Ahrefs: Google Seems More Biased Towards Big Brands Than ChatGPT and Perplexity (July 2025)
- Google Search Central: Optimizing for generative AI features on Google Search
- OpenAI Help: Searching the web with ChatGPT
- OpenAI: Overview of OpenAI Crawlers
- Search Engine Journal: GA4's AI Assistant Channel Undercounts Your AI Traffic (July 2026)
