Writing Product FAQs That AI Shopping Assistants Cite

The FAQ Went From SEO Tactic to AI Answer Source

For years, product FAQs were a mild SEO play. You bolted a few questions onto a product page, wrapped them in FAQPage schema, and Google might reward you with an expanded search snippet. Then, on May 7, 2026, Google deprecated FAQ rich results entirely. The expanded snippets are gone.

A lot of merchants read that headline and quietly deleted their FAQ sections. That was the wrong move. Google removed the rich result feature — the SERP appearance — not the value of the content or the schema. And the audience that now reads your FAQ answers most closely isn't Google's snippet generator. It's the AI shopping assistants your customers increasingly ask before they ever open a search results page.

When a shopper asks ChatGPT "is this jacket warm enough for winter hiking?" or asks Perplexity "does this brand offer free returns?", the model isn't inventing an answer. It's retrieving passages from the open web and synthesizing them. A well-written FAQ answer is one of the cleanest, most extractable passages you can hand it.

How AI Engines Actually Read a FAQ

The mechanism matters, because it dictates how you should write. AI search engines like ChatGPT Search, Perplexity, and Google's AI Mode use retrieval-augmented generation. Instead of ranking whole pages, they:

Chunk web content into passages — typically short, self-contained blocks of text.
Score each passage for relevance to the exact question asked.
Synthesize an answer from the highest-scoring passages, often citing the sources.

The critical implication: your FAQ answer is not evaluated as part of a page. It's evaluated as a standalone passage, usually somewhere between 40 and 200 words, competing against every other passage on the web that addresses the same question. This is a completely different game from traditional SEO, where the whole page's authority carried a thin answer along with it.

That reframes what a good FAQ answer looks like. It has to make sense with zero surrounding context, answer one specific question completely, and read like a fact the model can lift and attribute. A vague, keyword-stuffed paragraph that depended on the rest of the page to make sense is now dead weight.

What Questions to Actually Answer

The best product FAQ isn't a list of questions you wish customers would ask. It's a mirror of the questions they actually type into AI assistants — which are longer, more conversational, and more specific than old search keywords.

Mine real customer language. Your support tickets, live chat logs, and product reviews are full of the exact phrasing shoppers use. "Will this fit a queen mattress that's 14 inches deep?" is a real question. "Bed sheet sizing" is a keyword nobody says out loud to Gemini.
Cover the decision-blocking questions. These are the doubts that stop a purchase: sizing and fit, material and durability, compatibility, care instructions, shipping timelines, and return conditions. If a model can't answer these from your content, it either stays silent about you or pulls the answer from a third-party source you don't control.
Answer the comparisons. "Is X waterproof or just water-resistant?" and "What's the difference between the standard and pro version?" are the questions AI assistants love, because comparison is what shoppers ask them to do. Answering these directly is closely related to the comparison content AI engines love to cite.
Don't skip the unflattering ones. "Does this run small?" answered honestly builds the trust models reward. Ducking it just sends the model to a review site that answers it less kindly.

Structure Each Answer as a Self-Contained Unit

Once you know the questions, the writing itself follows a few rules that map directly onto how RAG systems chunk and score text.

Front-load the answer. Give the direct answer in the first sentence, then elaborate. "Yes, the jacket is rated for temperatures down to -10°C" beats a paragraph that builds to the answer at the end. Models — and impatient shoppers — reward the passage that resolves the question immediately.
Repeat the key noun; don't rely on pronouns. Because the passage may be extracted alone, "The blender's motor is rated for daily use" is safer than "It's rated for daily use." The model shouldn't have to guess what "it" refers to.
Keep each answer roughly 40–120 words. Long enough to be complete, short enough to be a clean chunk. If an answer sprawls past 200 words, it's probably two questions pretending to be one. Split it.
Be specific and verifiable. Concrete facts — measurements, materials, timeframes, compatibility lists — are what make an answer extractable and citable. Do not invent numbers to sound authoritative; a wrong spec that a customer catches destroys the trust that earns citations in the first place.
Write in plain, declarative sentences. No marketing throat-clearing, no "we're proud to offer." State the fact.

This is the same discipline that governs writing product descriptions AI engines can parse — clarity and structure over persuasion.

Keep the Schema, Even Without Rich Results

FAQPage schema no longer earns you a Google snippet. Keep it anyway. As Schema.org markup it's still valid, it won't hurt your pages, and it's still crawled by PerplexityBot, Bingbot, and the retrieval crawlers feeding AI answer engines. Structured data hands those systems your questions and answers as high-confidence, machine-readable pairs rather than making them infer the Q&A structure from your HTML.

A few rules that carry over from the schema markup fundamentals:

Only mark up FAQs that are visibly on the page. Hidden, schema-only FAQs are a guidelines violation and read as manipulation to AI crawlers too.
Keep Question.name and Answer.text in sync with the rendered content. Conflicting on-page and structured versions are a signal that gets you discounted.
Render the FAQ in HTML, not JavaScript a crawler never executes. If your FAQ widget only appears after a client-side script runs, many AI crawlers see nothing. This rendering gap is one of the most common issues a technical foundation review uncovers.

Where FAQs Belong

Placement is a strategic choice, not an afterthought.

Product-specific FAQs go on the product page, close to the questions they answer. This keeps the passage contextually anchored to the product entity the model is evaluating.
Policy questions — shipping windows, returns, warranty — can live on a dedicated page, but the highest-intent versions ("Can I return this specific item?") deserve a place on or near the product too. These overlap with the shipping and return trust signals AI engines read as proof of a legitimate merchant.
Category-level FAQs ("How do I choose a size?") belong on collection pages, where broader, discovery-stage questions get asked.

Scattering the same generic FAQ across every page helps nobody. Match the question's intent to the page's role in the buying journey.

The Bottom Line

The FAQ didn't die when Google dropped rich results — its audience changed. It's no longer a snippet tactic aimed at a search algorithm. It's a supply of clean, extractable, citable answers aimed at the AI assistants that now stand between your store and your customers.

Write each answer as a self-contained fact. Mirror how real shoppers phrase real questions. Be specific, be honest, keep the schema, and render it where crawlers can read it. Do that, and your FAQ becomes one of the most efficient pieces of AI visibility real estate you own. If you're not sure which questions AI engines currently answer about your products — and where they're pulling those answers from — an AI visibility audit is the place to start.