The 11% Problem — Why AI Platforms Recommend Different Businesses
Major AI platforms draw on different source mixes when answering local business questions — industry analyses find roughly 11% overlap in domains cited across engines for comparable prompts. Winning one assistant does not mean winning all; cross-platform measurement and broad signal coverage beat single-engine optimization.
One question, six different winners
A plumbing firm owner runs the same prompt in ChatGPT and Perplexity: "Best emergency plumber in Austin."
ChatGPT names a shop with 300 Google reviews. Perplexity cites a data-heavy local blog and names two different contractors. Gemini highlights a business with aggressive GBP posting. Claude mentions a third option with strong schema.
Same city. Same intent. Different answers.
This is the platform overlap problem — and it breaks strategies that treat "AI SEO" as a single score.
Where ~11% comes from
Multiple AI visibility vendors and research-oriented agencies have published cross-platform citation analyses. When they sample matched prompt sets across ChatGPT, Gemini, Perplexity, Claude, and other surfaces, they repeatedly find low shared-domain overlap — often cited in the ~10–15% range, with approximately 11% appearing in industry commentary as a shorthand.
Important caveat: This is not a peer-reviewed universal constant. Methods differ by:
- Prompt selection
- Geography and category
- Whether browsing is enabled
- Sample dates and model versions
Use 11% as strategic magnitude, not gospel. Your own free scan shows platform-specific mention tables for your market.
Why overlap is low
Different retrieval corpora
Perplexity emphasizes live web citations. Gemini integrates Google's index and local graph. ChatGPT blends training memory with optional browsing. Each stack surfaces different URLs for the same fact.
Entity vs document signals
Some engines lean on structured local entities (reviews, maps, directories). Others lift documents (articles, studies, listicles). A business dominant in entity graphs may be absent from document retrieval.
Model version drift
Weekly model updates shift weighting without public changelogs. Overlap percentages from Q1 2026 may not match Q3.
Prompt sensitivity
"Best plumber" vs "emergency plumber tonight" vs "who do neighbors recommend for slab leaks" pull different evidence. Aggregated overlap stats hide prompt-level variance.
Strategic implications for local businesses
1. Single-engine spot checks lie
A founder proud of a ChatGPT mention may still be invisible on Perplexity — where researchers and affluent buyers increasingly start.
Action: Six-platform baselines, monthly rescans. AI visibility tracking.
2. SEO wins do not generalize
Strong blue-link performance does not export to all assistants. Google's local pack strength helps Gemini; it does not automatically transfer to ChatGPT's browsing mix.
Action: Parallel AEO and GEO signal work — reviews, listings, entity, citations.
3. Coverage beats depth on one source
Investing only in Google reviews leaves gaps on engines that cite industry publications, data studies, or Apple Maps.
Action: Listings module across Google, Apple Business Connect, Bing, Yelp, vertical directories. Entity Profile with llms.txt and JSON-LD.
4. Competitor intelligence is per-platform
Your "#1 competitor" on ChatGPT may be irrelevant on Claude. Share-of-AI-voice tables must be platform-segmented.
A practical coverage matrix
| Signal | ChatGPT | Gemini | Perplexity | Claude |
|---|---|---|---|---|
| Google reviews | High | High | Medium | Medium |
| GBP freshness | Medium | High | Low–Med | Low |
| Apple BC / Siri graph | Medium | Low | Low | Low |
| Citable on-site data | Medium | Medium | High | High |
| Third-party press | Medium | Medium | High | Medium |
| Schema / llms.txt | Medium | Medium | High | High |
Cells are heuristic, not official weights — useful for prioritization workshops.
Technical foundation: llms.txt checklist.
How AIrecommend.ai operationalizes cross-platform work
Our scan samples six platforms and flags platform blind spots — engines where competitors appear and you do not.
Fixes map to Growth Engine modules without pretending one tweak fixes all assistants:
- Universal: Review Engine, Listings, Entity Profile
- Retrieval-heavy engines: Data Studies, Press Wire (Dominance)
- Google-heavy engines: GBP Autopilot (Dominance)
- Wrong cross-platform facts: AI Accuracy Repair (Dominance)
Super Pixel ties platform-agnostic AI traffic to booked jobs — because ultimately revenue matters more than citation overlap statistics.
What not to do
- Optimize for a single screenshot
- Assume GEO and AEO are interchangeable reporting products from different vendors without shared prompt methodology
- Buy "guaranteed multi-platform placement"
- Ignore Apple Business Connect because "customers use Google"
Measure your own overlap
Industry stats inform strategy; your data decides budget.
- Run the scan
- Export competitor mention tables per platform
- Note blind spots
- Fix universal signals first (NAP, reviews, entity)
- Add citable content for retrieval engines
- Resample in 30 days
Read also: How AI assistants choose businesses.
Bottom line
Low cross-platform overlap — often summarized as the ~11% problem — means AI visibility is inherently multi-engine. Treat anything that measures only one assistant as incomplete.
Honest programs report per-platform mention rates, refuse placement guarantees, and fix the broad signal stack you control.