FAQ Pages That Answer Engines Actually Quote

On May 7, 2026, FAQ rich results stopped appearing in Google Search. Google's documentation note was short and offered no reasoning, as Search Engine Journal reported: the search appearance, the rich result report, and Rich Results Test support all get dropped, with the Search Console API following in August.

Plenty of merchants read that and started deleting FAQ sections. That's the wrong move, and the reasoning behind it was never very good in the first place. If your FAQs existed to earn an accordion widget under a blue link, they were always a fragile asset. The reason to keep them has nothing to do with what Google draws on a results page.

The format survives because of how answers get built

Google describes its own process plainly in the AI features documentation: AI Overviews and AI Mode use a "query fan-out" technique, "issuing multiple related searches across subtopics and data sources" before composing a response. One shopper question becomes eight or ten machine questions. The answer gets stitched from whichever passages best resolve those sub-questions.

That changes what a "good page" means. Ahrefs re-ran its AI Overview citation study in March 2026 across 863,000 keyword SERPs and 4 million cited URLs and found that only 38% of cited pages also rank in the top 10 for the original query, down from roughly 76% in their July 2025 analysis. Around 37% of cited URLs don't rank in the top 100 at all. Ranking for the head term is no longer how you get pulled into the answer. Matching a sub-question is.

A page organized as discrete question-and-answer pairs is, structurally, a bag of pre-matched sub-answers. That's the entire advantage. It isn't magic and it isn't schema. It's that you already did the work of splitting one topic into the ten questions a machine was about to invent.

What a quotable answer looks like

Four rules, and the first one carries most of the weight.

The first sentence answers the question. No windup, no restating the question, no brand throat-clearing. If a model has to read three sentences to find out whether the answer is yes, it will find a competitor who said yes in four words.

Here's the version that doesn't get quoted:

"How long does the 8 oz candle burn? We take a lot of pride in our hand-poured soy blend, which is designed to give you a long, even burn and a beautiful scent throw in any room of your home."

And the version that does:

"How long does the 8 oz candle burn? About 45 hours, based on our in-house burn tests. Soy blend, single wick, trim to 5mm before each light."

One question, one topic. If your answer contains the word "also," you probably have two questions.

Facts over adjectives. Numbers, materials, dimensions, timeframes, compatibility. An assistant can repeat "45 hours" without any risk. It cannot repeat "beautiful scent throw" without sounding like an ad, so it won't.

Each answer stands alone. Assume it gets read with nothing above or below it, because that's exactly what happens.

Real questions, not keyword bait

The fastest way to write bad FAQs is to start from a keyword tool. You end up with the genre everyone recognizes: "What is the best organic cotton t-shirt?" answered by "Our organic cotton t-shirt is the best organic cotton t-shirt for quality and comfort." Nobody asked that. Nobody will ever quote it.

It's also the shape Google spells out under scaled content abuse, which it defines as generating many pages "for the primary purpose of manipulating search rankings and not helping users," providing "little to no value to users, no matter how it's created." The policy isn't anti-AI. It's anti-filler. Two hundred fake questions across a catalog is filler at scale.

The real questions are already sitting in your business. Search your support inbox for the last 90 days and filter for question marks. Pull pre-purchase chat transcripts, which are almost pure buying objections. Read your return reasons. Read your one and two-star reviews, where people explain what they expected and didn't get. In Search Console, filter queries containing "can," "does," "is," "how" and "vs." Twenty minutes across those five sources will give you more usable questions than a month of guessing.

Where FAQs belong on a Shopify store

Product pages first. The questions that stop a sale live here: fit, compatibility, materials, care, how long it lasts, what's in the box, whether it works with the other thing the shopper owns. Three to six per product, specific to that product.

Collection pages next, and this is the underused one. Collection FAQs answer category-level buying questions rather than product-level ones. What's the difference between the two finishes. Which one suits a small kitchen. How to choose a size without measuring. It also happens to be the cheapest fix for a collection page with an empty description field, which is a problem worth solving on its own terms if you're dealing with thin content.

Policy pages third. Shipping cutoffs, return windows, international duties, warranty terms. These are dull and they are the pages an assistant checks when it compares you to another store. If your return window is stated in a sentence a machine can parse, it can be quoted. If it's buried in a wall of legal text, it can't.

A central /pages/faq is fine as a catch-all, and it's the least valuable place you can put a question. Questions belong next to the thing they're about.

Keep the JSON-LD, and be honest about what it does

FAQPage is still a valid schema.org type, and Google's position is that you can leave existing markup in place. Unused structured data doesn't cause problems for Search.

What it won't do is buy you a citation. Google is unusually direct about this in the AI features docs: "You don't need to create new machine readable files, AI text files, or markup to appear in these features. There's also no special schema.org structured data that you need to add." Anyone selling you FAQ schema as a ticket into ChatGPT is guessing, and you should treat the claim accordingly.

So why keep it? Because it costs nothing, it's a clean machine-readable restatement of content you're publishing anyway, and it enforces a discipline that's genuinely useful: Google's rule that structured data must match what's visible on the page. If your markup and your page disagree, one of them is lying to somebody. The markup is the cheap insurance. The content is the asset.

Auditing what you have

Go template by template rather than page by page. Which templates render FAQs at all, and how many questions does each carry. Do the answers open with the answer, or with a paragraph of brand. Are the same three questions duplicated across 200 products, which is a sign they were bolted on rather than written. Does the visible text match the markup. And the one people skip: is any of it still true after last season's policy change.

Seokai generates FAQs per page and feeds them into FAQPage JSON-LD, and FAQ coverage is one of the inputs to the AI Readiness score, which is a reasonable proxy for "how much of my store can be answered from." The broader picture of what makes a catalog legible to assistants is in our guide on getting cited by ChatGPT, Gemini and Perplexity.

The widget is gone. The format outlived it, because the format was never really for Google. It was for anyone, human or machine, who arrived at your store with a question and thirty seconds of patience.

Point Seokai's site audit at your store if you want a list of which pages currently answer nothing at all.