From Short Query to Detailed Prompt: How Your Prospects Now Formulate Their Needs
The gist: When a prospect types a query into ChatGPT, the AI breaks it down into 6 to 12 sub-questions. If your content doesn’t answer any of them, you’re invisible. And these sub-questions are much longer and more precise than the keywords we used to optimize for.
What you’ll learn:
- Why your prospects no longer type keywords but detailed prompts
- What query fan-out is and how it decides who gets cited by AIs
- The 3 concrete changes to make to your content to appear in AI answers
- Why good SEO is still essential but no longer sufficient
Before you continue: This article is for you if you already have a website, track your Google rankings, but notice that ChatGPT, Perplexity, or AI Mode never mention you. If you’re not even indexed on Google, start with the SEO basics first.
Published June 30, 2026
“I’m looking for a plumber in the 3rd arrondissement of Lyon who can do emergency hot water repairs, with a travel fee included, and preferably with recent customer reviews.”
You would never have typed this into Google two years ago. You would have written: “plumber emergency Lyon 3”. That was the search engine constraint: condense your need into 3–4 keywords and hope for a result.
Today, that same sentence is natural in ChatGPT, Google AI Mode, or Perplexity. And it’s a silent earthquake for anyone selling online.
The problem: You’re still optimizing your pages for the short queries nobody types into conversational AIs anymore.
The solution: Understand how AIs turn a query into hidden questions — and answer them with your content.
The proof: Ahrefs’ massive study of 1.4 million ChatGPT prompts demonstrates that the AI citation mechanism relies on sub-questions nobody sees — except those who know where to look (Ahrefs, June 2026). On my end, I’ve been helping SMBs double their AI visibility in 8 weeks with this approach.
Why Your Clients No Longer Type Keywords
Martin Splitt and Nikola Todorovic (Google) confirmed it a few weeks ago: a new wave of users searches in a radically different way. More complex, longer, more conversational queries. They call this transformation “revolutionary” compared to search 10 years ago (Google Search Central, June 2026).
This isn’t a fringe phenomenon. Google AI Mode already reaches 1 billion monthly active users, and queries are doubling every quarter. AI Overviews appear in roughly 15% of searches. And according to a June 2026 Pew Research study, 60% of US adults have read an AI Overview, 50% use chatbots, and 25% use them daily (Pew Research, June 2026).
What I’ve learned in the field: A baker client in Montpellier told me, “I check ChatGPT before Google now — I describe what I feel like eating and see what it suggests.” He doesn’t even visit SERPs anymore. For him, the question isn’t “who’s #1 on Google?” — it’s “who will ChatGPT recommend?”
The difference isn’t trivial. Where Google forced you to decode the intent behind 3 keywords, AIs receive a complete intention, worded in natural language. This means your competition isn’t just the sites better ranked than yours — it’s every piece of content that better answers the exact question your prospect asked.
What this changes for you: If you’re still writing pages for “plumber Lyon”, you’re missing the other 40 possible formulations of the same need. And that’s exactly what Liz Reid (Google) calls “keyword fragmentation” — queries are becoming more varied and specific, making the traditional keyword research model obsolete (Google, May 2026).
How Does an AI Transform a Short Query Into a Detailed Prompt?
This is the most important mechanism to understand: query fan-out.
When a user types “best restaurant Bordeaux terrace sea view” into ChatGPT or Google AI Mode, the AI doesn’t search for that exact phrase. It breaks it down into sub-questions:
- What are the best restaurants in Bordeaux?
- Which ones have a terrace with a sea view?
- What do customers say about these restaurants?
- What’s the average price?
- Do I need to make a reservation?
- What’s the Google rating for these restaurants?
- etc.
Each of these sub-questions is a mini-search. The AI fetches sources for each sub-question, then merges everything into a single answer.
Semrush documented this mechanism in detail: query fan-out is the process by which Google AI Mode and ChatGPT break down a query, collect information, and merge results (Semrush, 2026). Marie Haynes explains that, concretely, instead of a single search, the system runs several in parallel.
What this means for you: Your page about “best restaurants in Bordeaux” can be excellent and well-ranked. But if it doesn’t answer “does restaurant X have a sea view?” — that precise sub-question — the AI will look for another source for that information. And that source, not yours, will be cited in the final answer.
Tip: Take your industry’s main query and type it into ChatGPT. Look at the follow-up questions it asks you. Those are its disguised fan-out queries. Write them down. Those are the real keywords to target now.
What Does Query Fan-Out Actually Look Like in the Numbers?
The Ahrefs study of 1.4 million ChatGPT 5.2 prompts is revealing (Ahrefs, June 2026). Here’s what it shows:
| Metric | Cited pages | Non-cited pages |
|---|---|---|
| Cosine similarity between title and max-matched fan-out query | 0.656 | 0.484 |
| Citation rate for natural language titles | 89.78% | 81.11% |
| Median age of cited pages | ~500 days | — |
| URLs cited out of 33 retrieved per prompt | ~16.5 | ~16.5 discarded |
The two numbers that change everything:
-
0.656 vs 0.484 cosine similarity — Your page title needs to resemble the sub-question ChatGPT asks itself internally. Not the user’s main query. It’s a subtle but fundamental distinction.
-
16.5 URLs cited out of 33 retrieved — ChatGPT retrieves about 33 URLs per prompt but only cites half. The other 16.5 are discarded based solely on title and snippet. Your title is the gateway.
The same pattern appears in Google AI Mode: an SE Ranking study of 10,000 keywords and 120,000+ citations showed that only 9.2% of URLs are reused from one query to another, and 21.2% of keywords had zero reused sources. The volatility is extreme (SE Ranking, June 2026).
Thomas Capper (Moz) confirms: the overlap between the organic top 10 and AI Mode citations is only 12% (Moz, May 2026). Being #1 on Google guarantees nothing in AI search.
How to Write to Answer the Questions ChatGPT Asks Itself
The pattern is simple but requires a shift in thinking. Instead of asking “what keyword do people type?”, ask yourself “what precise question will ChatGPT ask itself when breaking down the customer’s need?”
The Fan-Out Title Rule
If your prospect types “SEO agency for furniture manufacturing SMBs Lyon”, the AI’s internal sub-questions might be:
- “What is the best SEO agency for SMBs in Lyon?”
- “How much does SEO cost for an industrial SMB?”
- “What results can a furniture manufacturer expect from local SEO?”
- “What are the client reviews for agency X?”
Your H2 titles should be those questions, not “SEO for industrial SMBs Lyon”.
The Fan-Out Response Structure
First paragraph of each section: the direct answer. The AI scans the beginning of your content to decide whether to cite your page. If it has to read 3 paragraphs before finding the answer, it moves on to the next source.
The Ahrefs study shows that pages with natural language titles perform at 89.78% citation versus 81.11% for classic SEO titles. That’s an 8.67-point gap — enormous at web scale.
| Title format | Citation rate |
|---|---|
| Natural question (What is…) | 89.78% |
| Classic SEO-optimized title (Best X in Y) | 81.11% |
| Generic title (X Services) | ~75% |
What I’ve learned in the field: I tested this pattern on a locksmith client’s site in Nantes. We replaced his standard service pages (“Locksmith Nantes — fast emergency service”) with pages answering specific questions (“How much does a lock change cost in Nantes in 2026?”, “How to open a slammed door without breaking the lock in Nantes?”). Result after 6 weeks: ChatGPT cites him on 4 emergency queries, and Google AI Mode recommends him in the area. His Google SEO also improved, because both systems share the same ranking signals.
How to Actually Optimize Your Site for This New Way of Searching
Three actions, in order:
1. Identify Your Industry’s Fan-Out Queries (1 hour)
Take the 5 main queries that bring you the most clients. Type them into ChatGPT, AI Mode, Perplexity. Note EVERY suggestion, follow-up question, and variation they propose.
Complete this with Tom Capper’s (Moz) budget-friendly prompt tracking method: use Google’s “People Also Ask” enriched by ChatGPT to get your sector’s real fan-out queries.
2. Structure Your Content in Question-Answer Blocks
For each service page:
- Turn the H1 into a natural question the client would type into ChatGPT
- Turn each H2 into a precise sub-question (no generic headings)
- First paragraph of each section = immediate answer, no fluff
- Add a structured FAQ with real questions (not filler)
- Use list + table format for comparisons and pricing
3. Cover the Topic in Depth, Not Width
Semrush identified 6 pillars of AI-oriented SEO: thematic authority, entity clarity, structured content, freshness, technical accessibility, trust signals (Semrush, 2026).
Specific depth beats broad coverage. A page that explores “plumbing prices in Lyon in 2026” in detail will be cited more than a generic “plumbing services” page that skims 20 topics.
Tip: Use structured FAQ schema (FAQPage JSON-LD). AIs recognize it and use it to extract answers. Ahrefs tested it and results are mixed (no guaranteed boost), but combined with deep content, it’s an extra signal.
Recap checklist:
| # | Action | Done? |
|---|---|---|
| 1 | Identify my industry’s fan-out queries in ChatGPT | ☐ |
| 2 | Turn H1s and H2s into natural questions | ☐ |
| 3 | First paragraph of each section = immediate answer | ☐ |
| 4 | Add a structured FAQ with real questions | ☐ |
| 5 | Check that each page covers one topic in depth | ☐ |
| 6 | Test after 1 month: is my site cited in AI answers? | ☐ |
| 7 | Adapt other site pages to the same format | ☐ |
Score interpretation:
- 0–2: You’re still on the “short keyword” model. Start with step 1.
- 3–5: You have the basics. Keep enriching each page with Q&A.
- 6–7: You’ve made the turn. Monitor trends — fan-out queries evolve every month.
What Types of Content Are Most Cited by AIs?
The Ahrefs study gives a precise answer: cited pages have a median age of 500 days. Freshness alone isn’t enough — it’s semantic relevance that counts.
The over-represented formats in AI citations:
- FAQs and explainers: Q&A structured content
- How-to guides: Step-by-step walkthroughs
- Informational listicles: Lists with context and detail
- Comparison pages: Tables, alternatives, costs
- Deep dives: Focus on one specific topic, not an overview
Google AI Mode cites an average of 12.6 links per answer, but 90.8% are in sidebar blocks — not inline in the response. The most cited domains? YouTube (20.9%), Reddit (19.6%), and authority sites like Wikipedia. For local businesses, the Google Business Profile is the most effective lever for getting cited in AI Mode — Google cites itself via Maps (SE Ranking, June 2026).
Is SEO Dead? (No, and It’s Not Even Close)
Google officially confirmed in May 2026 that AI Overviews and AI Mode use the same ranking systems as traditional search. Gary Illyes (Google) is clear — same crawling, same indexing, same signals (Google, May 2026).
Lily Ray (Amsive) rightly calls out the hype cycle around every acronym announcing “the death of SEO” each time. Tom Capper (Moz) confirms: “GEO doesn’t replace SEO, SEO is the best starting point for GEO” (Moz, 2026).
What query fan-out changes is an additional optimization layer:
- Foundation: Technical SEO (indexing, speed, mobile) → essential
- Authority: EEAT, backlinks, trust signals → essential
- GEO: Answering fan-out queries, Q&A format, specific depth → new lever
Without steps 1 and 2, step 3 is useless. But without step 3, your site can perform well on Google while being invisible in ChatGPT.
Key Takeaways
- Query fan-out is the new ranking mechanism — ChatGPT and AI Mode break every query into sub-questions and only cite pages that answer them precisely.
- Your titles are the gateway — cosine similarity between your title and the fan-out query decides whether you get cited or discarded.
- SEO remains the foundation, GEO is the lever — without technical SEO, fan-out content won’t even be indexed. But with both, you cover Google + AI.
- The shift is underway — 1 billion AI Mode users, 60% of US adults reading AI Overviews, queries doubling every quarter. Those who adapt now will have a 6 to 12-month lead.
Next time you write a service page, don’t ask “what keyword do people type?” Ask yourself: “what question will ChatGPT ask itself when reading my client’s query?” The answer to that question is your page’s real title.
Go Further
- How to Make My Brand Appear in ChatGPT — step-by-step method to get cited by LLMs
- How AI Recommends Your Business (GEO) — the fundamentals of Generative Engine Optimization
- ChatGPT Generated Content: Will Google Penalize You? — what Google really says about AI content
- How to Find Keywords That Bring Real Clients — the keyword research method that still works
Keywords aren’t dead. They’re just hiding in the questions nobody sees.
Need local support? Discover our service areas across France.