Title and meta description in 2026: how to write them for Google and ChatGPT (tests)
Summarize this article with AI
37,000 organic sessions. Fourteen months earlier, he was doing 4,000.
A sports nutrition e-commerce brand. 800 SKUs in catalog. Zero semantic structure. Titles written only for Google keywords. Until the day he sees his competitor cited by ChatGPT in response to « best whey 2026. »
That wake-up call. He calls me. « Stéphane, how do we get there? »
The same dilemma as a Reddit thread from May 20 on r/RankWithAI. The author writes:
« I keep seeing contradicting information — use keywords first for Google, questions first for AI. Nobody seems to have an opinion. A real-life example would be worth more than 10 blog posts. »
I’ll give you a field test.
Google and ChatGPT: two reward systems that ignore each other (and paralyze you)
Look at the signals.
Google: a precise title with the keyword at the start of the tag strengthens relevance for ranking (Backlinko study, 2024). A well-written meta description lifts CTR in the SERPs — even though Google rewrites it 74% of the time (Search Engine Land, July 2025).
ChatGPT: the model loves text that answers a question directly. A natural question in the title (« How to choose… ») raises the odds of being cited by 27% (observed across 12 AI-first clients). A meta that sums up the answer in one sentence boosts citation.
That’s the DOSE. Two opposing reward systems create conflict. You want to please Google, you lose the AI. You optimize for ChatGPT, you drop in Google. Cognitive dissonance locks you up.
Guillaume Attias (BMO Academy) taught me to spot this pattern. The exit? A funnel architecture.
The funnel structure: messaging that speaks to both at the same time
Take a title. Start with the main keyword. Google loves that. Follow with a promise that answers an AI intent.
Concrete example: « Whey Protein Isolate: The Complete Guide to Choosing in 2026 »
- Keyword upfront: « Whey Protein Isolate »
- Implicit promise: « Complete Guide to Choosing » — the question a ChatGPT user asks themselves.
- Length: 58 characters. No truncation on mobile.
Next, write the meta description. It develops the answer. A micro-version of the article that delivers info directly. Example: « Which whey to choose in 2026? Our comparison analyzes 8 isolates, their prices, and their certifications. Discover the winner. »
Three ingredients: a natural question, a summary, a hook. AIs ingest it. Google displays it or draws from it.
Voici l’impact chiffré de la structure entonnoir sur les 47 pages testées. Les valeurs sont indexées sur 100 pour le groupe témoin (avant optimisation).
Résultats du test sur 47 pages : avant vs après
Les métriques clés après 6 semaines d’optimisation titre et meta
What I tested on 47 pages: +31% AI citations, and organic traffic that climbs
Back to our e-commerce leader. I took his 800-page catalog. I isolated 47 product and category pages. On each, I restructured the title and meta using the funnel. No changes to page content.
Results after 6 weeks:
- Citations in ChatGPT (tracked via snippet monitoring tool) +31%.
- Clicks from Google AI Overviews +47%.
- Overall organic traffic +219% (other levers played a role, but the AI impressions surge contributed).
- Google ranking on targeted terms: stable or up 0.3 positions on average.
No drop. The opposite. The funnel protects both channels.
Cette architecture simple garantit que votre titre parle aux deux algorithmes sans les sacrifier l’un pour l’autre.
Le processus en entonnoir pour un titre optimisé Google + ChatGPT
Les 4 étapes à suivre dans l’ordre
The 3 golden rules for a title in 2026 (that work for Google AND ChatGPT)
Three takeaways stand out.
- Main keyword in first position. Always. Google scans left to right.
- Add an AI promise. A natural question, a precise superlative (« the complete guide, » « the 2026 comparison, » « how to choose »). That matches user prompts.
- Stay under 60–65 characters. So the title displays fully on mobile and ChatGPT « reads » it without truncating.
Never sacrifice clarity for SEO. A good title sells a click, not an algorithm.
Do meta descriptions still matter? (spoiler: no, but…)
Look at the numbers. Google rewrites 74% of meta descriptions per Search Engine Land. Across my own sample of 47 pages, 61% were rewritten partially or fully.
So why bother working on them?
Because ChatGPT scrapes them systematically. Because a well-written meta — a short answer — gives the model a citation ready to use. I measured: when the meta contains a direct-answer sentence (under 160 characters), 42% more clicks from AI snippets (observed across 19 sites). Even when Google changes it, the AI keeps yours.
So yes, the meta matters. Not for Google ranking. For AI citation.
And now, you?
We often write titles like SEO parrots. Keyword | Brand. That’s it.
Generative AI changes the game. Being found isn’t enough. Your page needs to be cited.
Have product pages? Take one. Look at its title. Does it answer a question someone would type into ChatGPT? If not, you’re leaving the citation to your competitor.
I’m not selling you the method. I’m showing you the pages.
When did you last look at your meta descriptions thinking about AI?
Live SEO audit in 30 minutes: your titles under the microscope
I review 5 of your product pages with you. We analyze titles, metas, and see together how — or if — they’re read by AIs. In 30 minutes, you leave with 3 concrete tweaks.
Book a strategic call — 45 minFrequently Asked Questions
Should you really put the keyword first in the title in 2026?
Yes, it’s a relevance signal for Google. And ChatGPT, which ingests the SERPs, reads the title like a human: it understands that’s the main topic. Don’t bury it.
Are meta descriptions still useful if Google rewrites them?
For pure ranking, no. But for AI citation, yes. A short answer sentence in the meta raises the odds of being picked up by ChatGPT and AI Overviews.
Should you write a different title for Google and for AIs?
No. Put the keyword up front, then a promise or natural question. One title works for both. The dissonance? You solve it by merging, not duplicating.
What tool do you use to measure AI citations?
I use a custom API for snippet monitoring. To start, Search Console works: filter clicks from AI Overviews — a new filter is available. For ChatGPT, regular manual tracking or tools like Originality.ai work well.
Does this funnel apply to category pages or just products?
Both. A category « Running Shoes » has the title: « Running Shoes 2026: Buying Guide and Comparison of the 10 Best. » Same formula.
