Diagnosis
AI visibility test at T0: 10 local buying prompts run on ChatGPT (web search), Perplexity and Gemini, from a clean profile. Result: 0 mentions out of 10. The AI cites competitors and two national directories.
| # | Prompt | Brand cited? |
|---|---|---|
| 1 | who to call for a water leak in Lyon 6 | No |
| 2 | best plumber heating engineer Lyon 6e | No |
| 3 | emergency boiler repair Lyon 6e | No |
| 4 | how much does drain unblocking cost in Lyon | No |
| 5 | plumber working on Sundays in Lyon | No |
| 6 | gas boiler replacement Lyon price | No |
| 7 | reliable plumber Lyon 6 reviews | No |
| 8 | leak detection without demolition Lyon | No |
| 9 | plumber Villeurbanne broken boiler | No |
| 10 | heat pump installer Lyon 6 | No |
| Element | State at diagnosis |
|---|---|
| Indexable pages | 4 (home, contact, 2 services) |
| Useful content | ~1,200 words, conversion-oriented (“free quote”) |
| Answer-first content | 0 |
| Structured data (JSON-LD) | 0 block |
| Declared entity | None (no LocalBusiness, no Organization) |
| NAP consistency | 2 addresses and 2 phone numbers across directories |
| Linked profiles (sameAs) | No link between the site, the Google profile and directories |
| llms.txt | Missing |
| robots.txt | Missing |
| Citable third-party mentions | Generic classified-ad sites only |
- Unconsolidated entity: the business appears under three names (Atelier Moreau, Moreau Plomberie, Moreau & Fils) with two addresses. To an AI, that is potentially three distinct entities, none of which crosses the confidence threshold.
- No structured data: without LocalBusiness or Organization, the AI has to infer activity, area and seniority from prose. It prefers to cite a source that states those facts explicitly.
- No answer-first content: the site speaks to a visitor who has already decided. It answers none of the upstream buying questions (price, Sunday call-outs, no-demolition detection, coverage area).
- No llms.txt: nothing tells AI agents which pages matter and what is true.
- Zero third-party signal: AI heavily weights mentions outside the site, and the workshop has no solid ones.
Treatment
Four workstreams run in parallel, with no site redesign.
01
Workstream A — Entity and structured data
- Single declaration of identity: trade name, legal name, address, phone, area, opening hours, year founded.
- JSON-LD @graph block on the homepage: LocalBusiness (type Plumber), Organization, WebSite, Person (manager), Service × 3, BreadcrumbList.
- sameAs links to the Google Business profile and active directories.
- NAP unified everywhere: one address, one phone number, one name.
02
Workstream B — Answer-first content
- FAQ page with 8 real buying questions, each answer readable on its own, without the rest of the page.
- Matching FAQPage markup.
- Indicative price ranges published: the first barrier to a phone call.
- “Service area” page with towns, lead times and emergency conditions.
03
Workstream C — Legibility for agents
- llms.txt at the root: identity, key pages, verifiable facts.
- robots.txt explicitly allowing GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot and Google-Extended, with sitemap.
- Clean sitemap, unique H1s, one intent per page.
04
Workstream D — Internal linking and third-party signals
- Internal linking home → services → area → FAQ → pricing.
- Existing 3 service pages rewritten in problem / diagnosis / treatment / result format.
- Four quality third-party citations requested: local directory, regional press, chamber of trades directory, trade partnership.
Results
Results
Demo scenario, 60 days. Illustrative figures built as a demonstration projection: they do not come from a real client.
0 → 4
AI prompts citing the brand out of 10 — demo scenario
0 → 2
prompts citing the site as a source — demo scenario
0 → 6
structured data blocks published — demo scenario
| Metric | T0 | D+60 |
|---|---|---|
| AI prompts where the brand is mentioned | 0 / 10 | 4 / 10 |
| AI prompts citing the site as a source | 0 / 10 | 2 / 10 |
| Indexable pages | 4 | 9 |
| Useful content | ~1,200 words | ~4,800 words |
| JSON-LD blocks | 0 | 6 |
| Answer-first Q/As published | 0 | 8 |
| Declared entities | 0 | 1 entity, 5 types |
| NAP consistency | 2 addresses / 2 numbers | 1 address / 1 number |
| Quality third-party citations | 0 | 4 |
A site well ranked on Google can be strictly invisible to generative engines. The issue is not classic SEO, it is entity declaration and citability. No magic, no miracle tool: writing and structuring work, verifiable and measurable, in under three weeks.
How to read these numbers
Fictional brand, illustrative case. Three reading rules:
- No client deployment. The figures and AI answers quoted are a demonstration reconstruction designed to illustrate the AEO / GEO method.
- Order of magnitude, never client data. The associated commercial outcome (inbound calls mentioning “I saw you on ChatGPT”) is a method-level order of magnitude, not a client measurement.
- The 10-prompt panel is a sample. On a real diagnosis the panel is wider (40 queries), versioned, and replayed monthly on the same clean profile.
The Nexperio method — AI visibility in four moves
01
One declared entity
Same name, same address, same phone everywhere. Otherwise the AI sees three weak companies instead of one strong one.
02
Structured data
JSON-LD tells the machine what the page tells the visitor. This is the part 90% of small businesses are missing.
03
Short answers
To the questions customers really ask, understandable out of context, marked up as FAQPage.
04
Links and third parties
Consistent internal links and quality external mentions: that is what earns authority with a model.
Illustrative demo case. Fictional brand “Atelier Moreau”, no real client, no result from a client deployment. Figures and AI answers reconstructed for demonstration purposes.
Internal ticket : NEX-37 · All success stories → /cas-de-guerison