Optimizing every product page for two readers: human and AI agent
Summarize this article with AI
Two readers, one goal: the sale
A backpack manufacturer calls me one Tuesday morning. He has 8,400 product pages. A clean catalog. Competitive prices. And a drop in visibility on queries like « best 35 L backpack ».
The content wasn’t the issue. The reading was.
Google no longer ranks a page for a single display. Its AI features still rely on core search systems, but they change how the page is consumed. Microsoft, for its part, is evolving the Bing index from a list of pages into a source of facts. Result: a product page must please two readers. The human who buys. The AI agent who cites.
According to Ahrefs, optimizing for AI search means making pages easier for assistants to cite when users ask specific questions. It’s the continuation of what we already do.
Out of the manufacturer’s 8,400 pages, I audited 150 pages. Three findings. First, 23% of the pages had lingering 404 errors. Next, 41% offered no Product structured data. Finally, 67% answered zero related questions: warranty, dimensions, compatibility, nothing.
Two readers. One sale.
The AI agent doesn’t browse like a human. It scans. It extracts. It cross-checks consistency. If a data point is missing, it cites a competitor. If it’s vague, it ignores it.
The product page must therefore do two things at once: convert the human in three seconds, and give the agent facts to extract right away. Price, stock, weight, material, delivery time.
Nothing contradictory about that. One architecture, with two entry points. And nothing more.
At this client, we started by restructuring the product page URLs: removing useless parameters, adding breadcrumbs, syncing the displayed price with the structured data. Result after 30 days: AI citations doubled on the 50 pages we worked on. No paid campaign.
Ahrefs décompose l’optimisation IA d’une fiche produit en cinq étapes distinctes. Voici le pipeline complet à appliquer à chaque page, dans l’ordre.
Les 5 étapes d’optimisation d’une fiche produit
Le pipeline complet pour servir l’humain et l’agent IA, de la crawlabilité au suivi des citations
Crawlability: the starting point everyone forgets
If the AI agent can’t read the page, nothing else matters.
Google says it in the same terms as Ahrefs: AI features rely on core search systems. The page must remain crawlable and indexable. And useful, for humans and machines alike. It starts with technical access.
Three quick checks.
- Robots.txt: let AI bots through such as GPTBot, ClaudeBot, PerplexityBot, Bingbot. Some e-commerce sites still block them by reflex.
- Status codes: a product page returning 404, 499, or 5xx will never be cited. Ahrefs offers Site Audit and Bot Analytics to detect these errors.
- Cloudflare or equivalent: firewall rules can block AI crawlers without you seeing it in Search Console.
I observe a constant with my clients. Technical teams spend hours on internal linking. They forget that the AI agent uses the same routes as Googlebot. A broken route cuts off the citation.
A sports equipment client had 312 product pages returning 404 after a migration. The site worked for humans. The AI agent received dead pages. We redirected 312 URLs to the closest active pages. We resubmitted the sitemap. Within 21 days, AI citations came back on 87 pages. Not 312. 87. Why? Because a redirect isn’t enough. The destination page must also contain the useful facts.
Another point. A recurring 404 status can reveal a content opportunity. If an AI agent sends a human to a product URL that no longer exists, that’s a query to cover. Provided you create the missing page rather than leave it as an error.
Regular verification is the foundation. Without it, the rest doesn’t hold.
Two readers, one entry point.
The simplest approach is to simulate the AI agent. Take a product page. Ask a precise question about the product. Does the agent cite it? If not, test the URL with a fetch tool. Check the response code. Verify the presence of title and description metadata. This manual diagnosis takes 10 minutes per page. It’s worth doing on 20 key pages.
Write for the human who buys, not the algorithm that reads
The human reader wants a quick answer. The AI agent wants an exact answer. Both agree on clarity.
On a product page, information structure drives conversion.
- Title: product name, main differentiator, volume or format.
- First paragraph: concrete benefit in one sentence.
- Bullets: the 4 to 6 decisive features.
- Price and availability: visible without scrolling.
- Proof: a rating, a review, a « verified » mention.
A client selling trail running shoes rewrote 40 pages with this structure. His goal: reduce bounce rate. Not please AI agents. The bounce rate went from 61% to 47% on those 40 pages. And AI agents started citing the pages. Why? Because the answer to « what’s the weight of this shoe in size 42 » became extractable.
The question is no longer « which keyword to place ». It becomes « which question does my product solve ». Ahrefs advises creating content that matches AI search queries and is worth being retrieved. For a product page, that means answering the questions the AI agent can ask on your behalf: dimensions, compatibility, care, delivery time.
No need for 900 words. Three precise sentences are worth more.
Example. A power bank page. Instead of « Long-lasting battery », write « Charges an iPhone 15 from 0 to 100% in 1h40. Capacity of 10,000 mAh. Weight of 212 g. »
The human reader scans. The AI agent cites. You win.
Clarity.
Structured data: the product page speaks clearly
The facts must be easy to extract for AI. Schema.org markup is the common grammar.
For an e-commerce product page, three markup types do the job.
| Markup | What it tells the agent | Common mistake |
|---|---|---|
| Product | The content type: a product | Missing on variant pages |
| Offer | The price, availability, currency | Price out of sync with display |
| Review | The average rating and number of reviews | Aggregate rating without associated reviews |
A home appliance client added Offer markup to 127 pages. Before, the agent couldn’t read the price. After, AI answers mentioned the exact price on 31% of tested queries. Without changing a single comma of visible text.
Microsoft describes the evolution of the Bing index into a source of facts. Structured data is now the citation language. If your product page doesn’t expose the price in JSON-LD, the AI agent has to infer it. Sometimes it gets it wrong. Often it cites a competitor.
Markup lives in the code. The human reader can’t see it. The AI agent reads it first.
Markup must exactly reflect the visible content. Any mismatch creates an incoherent signal. Google and Bing ignore it.
Check alignment. Displayed price. Marked-up price. Displayed stock. Marked-up stock. Displayed weight. Marked-up weight. Full consistency.
I deploy this check on every e-commerce project. It takes 5 minutes per template. It eliminates 90% of inconsistencies.
How many of your product pages show full consistency between display and markup?
Freshness and consistency: the two signals AI tracks
The AI agent checks whether facts remain true. A product page that announces 12 units in stock while your ERP says 3 units loses the citation. Not because of ranking. Because of trust.
Ahrefs keeps this separate. Keeping facts fresh and consistent is a distinct step. For an e-commerce page, it touches four fields: price, availability, specifications, delivery time.
A client selling furniture changed the weight of a sofa. The visible page showed 48 kg. The Product markup still said 52 kg. The AI agent stopped citing it for 18 days. We synced it. The citation came back.
The AI agent is sending you a signal.
How do you keep pages fresh without spending hours on them?
- Generate JSON-LD markup from the same field as the display.
- Schedule a monthly check of the 20 most-viewed pages.
- When stock hits zero, update the offer immediately.
- When a price changes, check the validity date.
Price is the most sensitive field. An AI agent citing an outdated price sends a customer who expected the current price. Every update protects trust.
Google and Bing recalculate freshness continuously. A product page updated the same day gets cited more often than a page frozen for 12 months.
Consistency between pages matters too. If your product page states a 3-year warranty and the brand page states 2 years, the AI agent goes to the competitor who doesn’t contradict itself.
A consistency audit on 90 pages for a client revealed 14 conflicts. After correction, AI citations climbed. +40% on the pages involved.
Trust.
When was your last price update?
Measure what the AI agent cites (and what it ignores)
Optimizing a page doesn’t stop at publishing. Everything starts with the first citation.
Tracking comes in three layers.
- Access errors. Ahrefs Bot Analytics lists AI bot requests that fail. 404, 499, 5xx. Each error is a lost citation.
- Orphan pages. Ahrefs Web Analytics spots 404s that receive traffic from AI search. Redirect them to the closest active page.
- Citations in plain sight. Ask ChatGPT, Perplexity, or Copilot about the product queries you track. Note which brands come up. And which pages too.
A baby care client queried 36 AI requests once a week. He saw three pages cited consistently. They had three things in common: a compatibility table, an exact price in markup, a recent customer review. We duplicated that template on 22 other pages. Result: 17 new citations in 18 days.
This tracking requires no complicated tool. A spreadsheet. Ten minutes. Three times a month.
The AI agent reads before your prospect. Your product page must give it something to cite.
Everything I claim here, I’ve tested on real pages.
How many of your product pages does an AI agent cite today?
Audit your product pages for AI
A live 45-minute audit: bot access, data consistency, markup, conversion structure. You leave with the exact list of corrections to make.
Book a strategic call — 45 minFrequently Asked Questions
Do you need to rewrite all product pages for AI?
No. Start with the 20 to 30 pages that generate the most organic traffic. Fix the structure, markup, and access errors. The template comes after.
Which AI bots should you allow in robots.txt?
Allow GPTBot, ClaudeBot, PerplexityBot, Bingbot, and Google-Extended in your robots.txt. Also check your firewall and CDN: they can block these crawlers too.
Is Schema.org markup essential for AI search?
A citation is never guaranteed. On the other hand, facts stand out clearly. Without Product and Offer markup, the agent guesses. And when it guesses, it often cites a competitor.
How do you check whether an AI agent cites a product page?
I ask precise questions about the product to ChatGPT, Perplexity, and Copilot. I note the URLs cited. Then I rerun the test every week, on the same key queries.
Does data freshness really influence AI citations?
Yes. An outdated price, stock, or specification creates a conflict. The AI agent takes up-to-date sources. Sync every field with your ERP.
