ChatGPT recommends your competitors — and not you. The diagnosis.
When AI answer engines cite competitors and ignore you, it almost always comes down to one of five causes. Each one is testable, four of them today. The complete diagnostic protocol.
When ChatGPT, Perplexity or Google AI Overviews recommend competitors for your service and don’t mention you, it almost always has one of five diagnosable causes: blocked access, poor machine readability, no citable answers, an unclear company entity, or missing third-party sources. Each cause has its own test — and its own fix.
This post is the diagnostic protocol we use internally. You can run four of the five tests yourself today, without buying any tools.
First: how to ask the diagnostic question properly
Ask the engines the way your customers ask — not the way you wish they would. Three phrasings, each in ChatGPT (with search), Perplexity and Google (AI Overviews):
- “Recommend a [your service] in [your city]”
- “[A concrete question a customer asks before buying]”
- “What is the best provider of [service] for [audience]?”
Record per answer: who gets cited? With which page (homepage, blog post, directory listing)? Important: answers vary between runs. A single query is an indication, not a finding — which is why we measure monthly with a fixed question matrix, not with one-off tests. But for diagnosis the pattern is enough: if you appear zero times in ten queries and a competitor appears seven times, that is not chance.
Cause 1 — The door is shut: AI crawlers are locked out
The test (10 seconds): our AI access check reads your robots.txt live and shows, per crawler — GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended and five more — whether it may read your website.
The pattern: security plugins, CDN bot filters and robots.txt templates from when AI crawlers were new often block everything that isn’t called Googlebot. How websites lock out AI without knowing it — covered in detail here.
The fix: a per-vendor allowlist in robots.txt for the search and answer crawlers. An hour of work; re-ingestion begins afterwards — Perplexity typically reacts fastest in our experience, because it reads live instead of from an index.
Cause 2 — The door is open, but there’s nothing to read inside
AI crawlers are worse JavaScript renderers than Googlebot. If your page content only exists after four scripts have loaded, an answer engine sees an empty shell — and cites the competitor whose content is in the HTML.
The test (2 minutes): open your most important service page, then “View page source” (not the developer console — the source). Ctrl+F for a sentence that is visible on the page. If you can’t find it, your content doesn’t exist for many machines.
The fix: pre-rendered HTML — for us that means Astro and Vercel; on legacy systems at minimum server-side rendering. The next level: machine-readable Markdown versions of key pages plus an llms.txt as a curated table of contents — we published the full guide.
Cause 3 — Readable, but nothing is citable
Answer engines cite passages, not pages. A page full of “we are your competent partner for holistic solutions” contains not one sentence an AI could use as the answer to a concrete question.
The test (5 minutes): take the answer in which your competitor is cited and open the cited page. Almost always you’ll find: a direct answer in the first sentences of a section, concrete numbers, a question as a subheading. Compare that with your own page on the same topic.
The fix — the four ingredients of citable content:
- Answer first: every section opens with the 40–60-word answer, then the reasoning — not the other way round.
- Numbers with source and date. The Princeton GEO study (KDD 2024) quantified it: citing sources +40% visibility in AI answers, statistics +37%, quotations +30% — and keyword stuffing −10%.
- Questions as headings, phrased the way customers ask them.
- Tables for comparisons, numbered lists for processes — both get lifted whole by answer engines.
Plus clean schema markup — FAQ, Service and Article annotation, so machines don’t have to guess the structure.
Cause 4 — The AI isn’t sure who you are
Language models work with entities. If your company name appears sometimes with the legal suffix and sometimes without, your address and phone number differ between directories, and no Organization markup exists, your company is a cloud of fragments to the machine — and machines don’t recommend what they’re not sure about.
The test (1 minute): simply ask ChatGPT and Perplexity: “Who is [your company name] in [city]?” A precise, correct answer means: entity clear. Mix-ups, stale facts or “I have no information on that” means: cause found.
The fix: Organization schema with consistent master data on every page, identical name–address–phone data in every directory, linked profiles (sameAs: LinkedIn, Google Business Profile, industry directories) — the work we call brand entity management.
Cause 5 — You only exist on your own website
The most uncomfortable finding: analyses of AI citation patterns show brands are cited via third-party sources a multiple of times more often than via their own domain — directories, review platforms, trade articles, Wikipedia. If your competitor appears in three industry directories, one trade-magazine interview and twenty Google reviews while you exist only on your own site, the AI cites them — from sources you don’t control, but which pass judgement on you both.
The test: look closely at which pages are cited in your diagnostic answers from step one. Often it isn’t the competitor’s website at all — it’s a directory or portal they’re listed in and you aren’t.
The fix: build presence where the engines already read: a maintained Google Business Profile, the relevant industry directories, review workflows — and, medium-term, original data studies that journalists cite: the only link source that feeds classic SEO and AI citations at the same time.
The protocol at a glance
| # | Cause | Test | Time | Tool |
|---|---|---|---|---|
| 1 | Access blocked | AI access check | 10 sec | free |
| 2 | Content not in HTML | Source-code search | 2 min | browser |
| 3 | Nothing citable | Compare the cited competitor page | 5 min | browser |
| 4 | Entity unclear | Ask “who is [company]?“ | 1 min | ChatGPT/Perplexity |
| 5 | No third-party sources | Note cited sources, not brands | 10 min | browser |
What you don’t control — and what that means for measurement
Honesty is part of the diagnosis: a share of the outcome is not yours to determine. Models have training cutoffs, answers vary between runs, and providers change their source selection without notice. The consequence: measure trends, not single cases. For clients we run a fixed monthly question matrix per engine and track the movement — visible in the live dashboard, not in a monthly PDF.
Closing exactly that gap is a dedicated pillar at SEO216 — entity-building, citable structure and citation tracking across five AI engines.
Once you’ve run the protocol and want the findings prioritised and fixed: the free audit includes the citation analysis per engine, the comparison against your strongest competitor and a prioritised action plan — within 24 hours.
