You ask ChatGPT who the best providers in your category are. It names three companies. One of them is a competitor you genuinely believe you outperform — smaller team, thinner product, worse reviews. You are not mentioned at all.
The instinct at this point is to assume you need more content, more backlinks, more everything. Usually that is wrong, or at least premature. In most audits we run, the gap traces back to one specific asymmetry — and until you identify which one, extra effort mostly gets spent in the wrong place.
This guide is the diagnostic. Not a list of generic causes, but the actual comparison method: how to find out what your competitor is doing that you are not, engine by engine.
The short version
Start by checking whether your competitor is actually beating you everywhere or just on the one engine you happened to test — cross-engine citation overlap is surprisingly low. Then separate three different problems that look identical from the outside: not being in the answer, not being cited as a source, and not being credited by name. Each has a different cause and a different fix. The most common real cause is a sourcing deficit, not a content deficit.
Step ZeroAre They Actually Beating You, Or Did You Check One Engine?
This sounds like a technicality. It is the single most common measurement error in AI search, and it changes the entire diagnosis.
The engines do not cite the same sources. A 2026 per-engine audit compiled by Digital Applied found that only around 11% of domains cited by ChatGPT overlapped with domains cited by Perplexity. Same query category, almost entirely different source sets. That is not noise — it reflects genuinely different sourcing philosophies between the engines.
Even more counterintuitive: Google rankings do not reliably transfer. Semrush AI visibility analysis published in early 2026 found that only a small fraction of pages ranking in Google's top ten also appeared among ChatGPT's citations. Practitioners have started calling this the decoupling of Google rankings from AI citations, and it explains a lot of confused reporting. You can be page-one on Google and functionally invisible in ChatGPT.
So before concluding a competitor is beating you, check all four major surfaces. It is common to find that they dominate ChatGPT while you quietly own Perplexity, or that Google AI Overviews cite you and nobody noticed because nobody was looking at that report — Search Console reports AI Overview impressions if you know where to filter.
You cannot diagnose a gap you have measured on one engine. The engines barely agree with each other.
PrecisionThree Different Problems That Look Identical
"My competitor shows up and I don't" describes at least three separate failures. Conflating them is why so many fixes miss.
| What to measure | What it means | What it tells you |
|---|---|---|
| Share of answer | How often your brand appears in the body of the AI's response | Whether you are in the consideration set at all |
| Share of citation | How often your actual URL is credited as a source | Whether the engine trusts your site as evidence |
| Share of mention | How often your brand name is referenced anywhere | Whether your entity is recognized in the category |
The diagnostic value is in the ratios, not the absolute numbers. As the team at LLM Pulse put it in their 2026 measurement guide: high citations with low brand mentions means the AI is using your content but not attributing your brand. The inverse — mentioned often but rarely cited — signals a sourcing deficit. Two completely different problems, two completely different fixes.
Worked example. You track 250 prompts weekly across five engines. Your brand appears in 38 response bodies, is cited by URL in 19, and is mentioned somewhere in 62. Your nearest competitor scores materially higher on citation specifically. That single gap tells you where to spend — it is a sourcing problem that digital PR and third-party presence close, not a content-volume problem more blog posts will fix.
The MethodRun the Teardown in an Afternoon
You do not need a paid tool to do the first pass. You need a spreadsheet and about two hours.
- Write twenty real buyer prompts. Not keywords — full conversational questions with constraints, the way a buyer actually types them. Pull them from sales calls, support tickets, and your Search Console long-query data. If you want the full method for sourcing these, we broke it down in keyword research for AI search.
- Run each one through four engines — ChatGPT, Perplexity, Gemini, and Copilot — in fresh incognito windows so personalization does not skew results.
- Record four columns per prompt: is your brand in the answer body, is your URL cited, is your competitor in the answer body, and — the important one — which third-party sources did the answer pull from.
- Read that fourth column across all twenty rows. Patterns emerge fast. You will usually find the same handful of publications, review platforms, or directories appearing repeatedly. That is the source pool the engine trusts for your category.
- Check whether your competitor is in those sources and you are not. This is the answer to your question about 70% of the time.
The CausesFive Asymmetries That Decide Who Gets Named
They are in the sources the engine weights heavily. You are in sources it doesn't.
This is the biggest one, and it is not about volume. AI engines weight specific source types — established publications, category-specific review platforms, structured data providers, recognized authorities — far above general web content. Being mentioned in fifty low-authority blogs loses to being mentioned in three sources the engine actually trusts.
This is also why size does not protect you. The Drum reported in July 2026 on exactly this pattern: newer entrants dominating AI answers on a fraction of an incumbent's total media coverage, because they were consistently present in the specific high-weight sources for their category.
Their content answers the question. Yours ranks for the keyword.
AI engines extract passages, not pages. A competitor page that opens a section with a clean one-sentence answer gets lifted; your page that buries the same information under three paragraphs of positioning copy does not — even if yours ranks higher and is more thorough.
The Princeton and IIT Delhi GEO study (KDD 2024) tested nine content tactics across roughly 10,000 queries and found the strongest gains came from adding named citations, adding attributed statistics, and improving fluency — each delivering a 30-to-40% relative improvement. Keyword stuffing scored below the do-nothing baseline. If your competitor writes plainly and cites sources while you write keyword-dense marketing prose, they win the extraction regardless of rank.
Their brand is a recognized entity. Yours is a domain name.
AI engines need to know what you are before they can recommend you. A competitor with a Google Knowledge Panel, a Wikidata entry, consistent NAP data, and Organization schema with sameAs links is an identifiable entity in the model's world. A brand with none of those is a URL the model cannot confidently place in a category.
This asymmetry is particularly brutal for newer brands and for anyone whose brand name is also a common word, because the engine cannot disambiguate.
Their site is accessible to that engine's crawler. Yours quietly isn't.
Binary and boring, but we find it constantly. The engines use different crawlers — OAI-SearchBot for ChatGPT search, GPTBot for OpenAI training, plus PerplexityBot, ClaudeBot, and Google-Extended. Blocking one removes you from that engine entirely while leaving the others unaffected, which produces exactly the pattern of "invisible on one engine, fine on another" that confuses people.
Two hidden variants: a CDN or WAF bot rule that drops requests at the edge before robots.txt is consulted, and client-side rendering that leaves critical content invisible to crawlers that execute JavaScript less reliably than Googlebot.
They have the verification footprint. You are a risky recommendation.
AI assistants are conservative about recommendations because a bad one costs them credibility. Review volume, rating consistency across platforms, and the existence of independent comparison content all function as risk-reduction signals. A competitor with 400 reviews across three platforms is a safer name to output than you with 12 on one.
This compounds at the conversion stage too. BrightLocal's 2026 survey found that while 63% of active AI users trust AI business recommendations, 88% still fact-check before acting — most commonly by reading reviews. So a weak review footprint costs you the recommendation and then costs you again if you get one. For location-based businesses this overlaps heavily with local SEO for AI search, where review consistency across platforms carries even more weight.
PrioritizationWhich Gap to Attack First
Fix in speed order, because early wins fund patience for the slow work:
- Week 1 — crawler access. Free, fast, and it gates everything else. If an engine cannot fetch you, nothing downstream matters.
- Weeks 2 to 4 — entity and schema. Organization schema, sameAs links, directory consistency, and passage restructuring on your top ten pages.
- Month 2 onward — content restructuring at scale. Work through the pages that map to your teardown prompts, one direct answer at a time.
- Months 2 to 6 — the sourcing deficit. Pursue the specific high-weight sources your teardown surfaced. Slowest, hardest, most defensible.
- Ongoing — reviews. Start immediately; it accrues in the background.
A note on why this is worth the effort rather than a vanity exercise: being cited measurably pays. Seer Interactive, tracking 3,119 informational terms across 42 organizations, found brands cited inside an AI Overview earned 35% more organic clicks and 91% more paid clicks than uncited brands on the same queries. Presence in the answer is not a soft brand metric.
Honest CaveatsWhat This Analysis Can't Tell You
A few limits worth naming, because a lot of content in this space overclaims.
AI answers are non-deterministic. Run the same prompt twice and you may get different brands named. That means single-prompt observations are close to meaningless — you need a consistent set of prompts checked repeatedly over time before a pattern is real. Do not restructure your strategy around one unlucky query.
Measurement in this category is also young, and much of the published data comes from companies selling AI-visibility software. The share-of-voice figures and cross-engine overlap percentages cited above come from vendor analyses; the methodology is rarely fully disclosed. They are directionally useful and consistent with each other, which is why we cite them, but they are not peer-reviewed. The Princeton study and Seer's dataset are the firmer ground.
Finally: sometimes the competitor genuinely is better positioned, and the honest answer is that closing the gap takes quarters rather than weeks. If they have eight years of category authority and you launched last spring, the sourcing deficit is real and no technical fix shortcuts it. What you can do is win the narrower, more specific queries first — that is where incumbents are weakest, and it is how newer brands get a foothold. The broader diagnostic version of this sits in why your website isn't showing up in AI, and the systematic approach in generative engine optimization.
By IndustryWhich Asymmetry Usually Bites Hardest
The five asymmetries are universal, but their weighting is not. Patterns we see repeatedly:
- Regulated categories — in healthcare and banking and finance, asymmetry three (entity and credential clarity) dominates. AI engines apply visibly stricter standards to advice that affects health or money, so a competitor with named credentialed authors and verifiable licensing beats a better product with anonymous content.
- Local and location-based — in real estate and hospitality, asymmetry five (reviews and verification footprint) usually decides it, because AI leans on review platforms heavily for anything geographic.
- B2B and industrial — in construction and industrial and most SaaS categories, asymmetry one (the sourcing deficit) is the whole ballgame. Trade publication and directory presence decides who appears in procurement research.
- Ecommerce and product — attribute completeness and structured data outrank everything else, because AI shopping is constraint matching rather than keyword matching.
Run the teardown before assuming which one applies to you, though. The industry pattern is a prior, not a diagnosis.
Keep ReadingRelated Guides
- Why your website doesn't appear in ChatGPT — the ChatGPT-specific version, including the GPTBot and Bing-index angle.
- 10 strategies to increase AI search visibility — the tactical playbook once you know your gap.
- Google officially names GEO and AEO in 2026 — why the terminology in this article stopped being jargon.
- AI Overviews optimization — if your teardown shows Google's answer box is where you are losing.
We'll run the teardown for you
A free audit across ChatGPT, Perplexity, Gemini, and Copilot — with the specific sources your competitors appear in and you don't, plus a prioritized gap list.
Get Your Free AI Visibility AuditQuestions first? Talk to us. Or see the industries we work in.
Frequently asked questions
Why does ChatGPT recommend my competitor instead of me? +
Does ranking on Google mean I will show up in ChatGPT? +
Why do different AI engines recommend different brands? +
How do I find out why a competitor is winning AI citations? +
Can a smaller company outrank a bigger competitor in AI search? +
What is AI share of voice and how is it different from citations? +
How long does it take to close an AI visibility gap? +
Sources & references
- Aggarwal et al., Princeton / Georgia Tech / IIT Delhi / Allen Institute — "GEO: Generative Engine Optimization" (arXiv:2311.09735, ACM SIGKDD 2024) — the nine-tactic test; keyword stuffing underperforms baseline
- Seer Interactive — AIO Impact on Google CTR (3,119 terms, 42 organizations; cited brands +35% organic, +91% paid clicks)
- BrightLocal — 2026 Local Consumer Review Survey (63% trust AI recommendations; 88% still fact-check)
- Digital Applied — AI Share of Voice framework (per-engine overlap audit; Semrush AI visibility figures)
- LLM Pulse — Share of Voice in AI Search (share of answer vs citation vs mention distinction)
- Google Search Central — AI features and your website
- OpenAI — Crawler documentation (OAI-SearchBot vs GPTBot)
- Schema.org — Organization and sameAs reference
Sourcing note: the cross-engine overlap and share-of-voice figures in this article come from vendor-published analyses by companies that sell AI-visibility software. Their methodologies are not fully disclosed and none are peer-reviewed. We cite them because multiple independent vendors report consistent directional findings, but treat the specific percentages as indicative rather than precise. The Princeton study and Seer Interactive dataset are the more rigorously documented sources here.