Finding relevant keywords for LLMs

We're moving away from an SEO based on chasing traffic, as much traffic as possible to push up our curves in Search Console or Semrush. And that's over.

Today, we can see that SEO traffic has dropped, and what you need to look for is the semantic relevance of a keyword, not just its volume. Why? Because we know that with LLMs, a search is now made up of around 24 words, which completely changes the type of keywords we're going to target.

If queries used to be simple to find, today it's more complex, and classic tools aren't always enough, especially for long-tail queries. You have to build a specific workflow to be sure you identify the right queries and don't write for nothing.

What is a relevant keyword in an SEO strategy in 2026?

Before tackling the selection, first understand why 88% of SEO sites don't appear in AI. It's the blind spot most audits miss.

The way we research our keywords hasn't changed in 10 or 15 years: volume + difficulty. Except that the data from paid or free SEO tools is very often projected (not real), often based on English-language data, and so not necessarily on our French market.

On LLMs, the most-cited sources are Reddit, Wikipedia, YouTube, LinkedIn. On high-volume keywords, you have little chance. You need less visible, more expert keywords: micro-intents, long tail.

We've gone from 4 to 24 words for a search on LLMs. No traditional tool will give you a volume for these 24-word queries, yet added together, they can represent 100 potential customers.

A keyword with a volume of 10 or 20 can have a CPC (Cost Per Click) of 100 or 150. If you rely only on volume, you'll miss that huge business interest.

Criteria for selecting your SEO keywords

• Business interest: Does this keyword feed the business (leads/sales)?
• Semantic interest: Does it help strengthen your topical authority on your subject?
• CPC (cost per click): If the other players in your market are paying a lot, the keyword surely has business potential (email capture, demo, freemium, etc.).
• Volume: It remains an indicator, but a secondary one, mainly to limit the risk of getting the keyword choice wrong.

How do you find relevant keywords, step by step?

• Create themed bots: Create a bot on Gemini, ChatGPT or NotebookLM and feed it the relevant sources for your topic to give the bot context. A prompt without context is of no use. This will let you query the proprietary database you've just built and identify new relevant keywords.
• Use Google Keyword Planner: Use this tool first to get the "base data" and see whether the keyword has a CPC (real Google data).
• Find your keywords with Fusionn: It combines Google data and prompts to identify all the vectors around a keyword. Because for each main keyword, you need to be able to identify 5 to 20 secondary keywords (related intents). These micro-intents are what maximise your semantic score and help you climb on Google. It's also the basis of the grid method on the strategy side.

My SEO workflow:

1. Create themed bots (Gemini, GPT, Notebook LM) with context
2. Identify 1 to 5 (or 1 to 10) business keywords via Keyword Planner
3. Run my toolbox of operational AEO prompts in my context
4. Go through Fusionn to multiply the results
5. Create lists, select the best keywords
6. Validate with: search intent, funnel, business potential, email capture, page type, using a specific prompt

Ending "volume" as the indicator for choosing keywords

The traditional approach ("high volume = good keyword") is obsolete, and even dangerous for an LLM-specific strategy.

For three reasons:

• Skewed data: Most SEO tools use "projected data" (often English-language) rather than real data. There is a large gap with Google's reality.
• The volume / business paradox: A keyword with 500 searches can earn less than a keyword with 10 searches. Example: a keyword with 10 monthly searches can have a CPC (cost per click) of €100 to €200, which shows strong business intent, invisible to those who filter by volume alone.
• The "zero volume" opportunity: There are queries invisible to the tools (0 volume displayed) that nonetheless hide dozens of highly qualified potential customers, notably through micro-intents.

Think decision-stage and action queries

To exist against LLMs that "eat" informational content, you need to aim for the bottom of the funnel and for a level of expertise.

Types of queries to favour:

• Decision queries: the user arrives on the page to make a final decision (buy or not).
• Action queries: the user is looking to carry out a specific action (book a demo, use a simulator, take a call).

Finding the "Content Gaps"

These are the unexploited angles on your keyword. To be the SERP's originality, you have to cover what others don't dare to cover.

Example: if nobody gives their price in the plumber niche, and you do, you create an arbitrage. You become a "bonus result" that Google likes.