How to write FAQs so AI engines actually cite them

Shipping FAQPage JSON-LD puts your content into the retrieval pool, but the schema itself does not determine whether you get cited — the writing inside the schema does. AI engines extract FAQ answer text almost verbatim when they cite you, which means every word of the answer is copy that will be read by a real buyer deciding whether to trust your store. Treat FAQ content like the landing page it actually is. Generic, promotional, or vague answers perform worse than no FAQ at all because they dilute the schema signal without adding extractable value.

The writing discipline is what separates a FAQ that earns citations from one that just satisfies a schema validator. The six rules below are the ones we apply when we rewrite FAQ content for Shopify customers — they are not style preferences, they are specifically calibrated to how retrievers extract and rank passages.

1. Phrase the question the way a buyer would type it

'How long does shipping take?' beats 'What are your shipping policies?'. Match the phrasing a real shopper uses when asking an AI assistant — colloquial, direct, first-person if natural.

2. Answer in 40-120 words

Under 40 words and you are too thin for a standalone quote. Over 120 and retrievers truncate mid-sentence, which drops citation weight. 60-80 is the sweet spot for most questions.

3. Anchor every answer to a specific fact

A number, a dimension, a named material, a timeframe, a currency. 'Ships in 3-5 business days via UPS Ground' is citable; 'Ships quickly' is not.

4. No promotional adjectives

'Premium', 'handcrafted', 'luxurious' are zero-signal words for retrievers. Replace with specifications. If you earned 'handcrafted', prove it — 'assembled by hand in our Portland workshop by a team of four watchmakers'.

5. Make each answer stand alone

Answers are quoted in isolation. 'It fits wrists from 6.5 to 8.5 inches' fails if there is no 'it'. Name the product or attribute explicitly in the answer text.

6. Match the answer content to reality

Retrievers cross-check FAQ answers against review text and support-email patterns. A FAQ that says 'ships in 2 days' when reviews complain about 10-day delays gets both demoted.

Almost every merchant we audit started their FAQ page from a hypothetical — 'what would a shopper probably want to know?' — and ended up with a generic FAQ that matches no one's actual phrasing. The faster path is the support inbox. Your customer-service team is already answering the real questions buyers actually have, in the words buyers actually use, at the frequency buyers actually ask them. Three months of support emails produces a ranked list that beats any brainstorming workshop.

1. 1Export the last three months of customer-service email subjects and first-response content (or the equivalent from your helpdesk tool — Gorgias, Zendesk, Re:amaze, Intercom).
2. 2Cluster the emails by topic. A simple manual pass on 200-500 emails is usually enough — tooling is optional.
3. 3Count the cluster frequencies. The top 8-12 clusters are almost always the 80% of inbound support volume.
4. 4For each top cluster, phrase a question the way a buyer would ask an AI assistant (rule 1 above), not the way your support team categorizes it internally.
5. 5Draft the answer from the best-written customer-service response in the cluster — the one the team would use as a template — and tighten it to 60-80 words.
6. 6Validate each FAQ pair against the six rules and publish to the metafield that drives your FAQPage schema.

Across 1200+ Shopify stores, the same five question families account for roughly 80% of buyer-intent AI queries. Your specific wording will vary — a skincare store and a watch store do not ask the same sizing question — but the categories repeat. Plan your FAQ around these five rather than trying to cover every edge case. Edge cases dilute the signal without adding citation lift.

The five families with example phrasings

Shipping + fulfillment

'How long does shipping take?', 'Do you ship to [country]?', 'How much is expedited shipping?', 'Can I track my order?'

Sizing, fit, and dimensions

'What size should I buy?', 'Does this run small or true to size?', 'What are the exact dimensions?', 'Does this fit wrists above 8 inches?'

Materials, ingredients, care

'What is this made of?', 'How do I care for this?', 'Is this hypoallergenic / waterproof / vegan?', 'How do I clean this?'

Warranty, returns, authenticity

'What warranty is included?', 'What is your return policy?', 'How do I verify this is authentic?', 'What if it arrives damaged?'

Use cases and fit-for-purpose

'Is this good for [occasion]?', 'Can I wear this for [activity]?', 'Is this suitable for [target persona]?', 'Can I use this with [adjacent product]?'

1. 1Marketing-speak in the answer. 'Our handcrafted timepiece embodies old-world elegance' is zero signal. Replace with specifications — movement type, case diameter, water resistance rating.
2. 2Generic question phrasing. 'What are your shipping policies?' is too broad. Split into 'How long does shipping take?' and 'Do you ship internationally?' — each gets its own citable answer.
3. 3Vague timelines or ranges that hide the truth. 'Shipping varies by location' is a non-answer. If your US ships in 3-5 days and international in 7-14 days, say both.
4. 4Specs buried in paragraph prose. A 200-word answer with the relevant dimension buried in sentence 4 performs worse than a 60-word answer that puts the dimension in sentence 1.
5. 5FAQ content that contradicts the product description, review content, or shipping policy. Retrievers cross-check and demote both. If your FAQ answer says 2-day shipping and your shipping policy page says 5-day, pick one and make them match.

FAQ citation is measurable, not mysterious. Set up a weekly prompt panel of the 20 questions in your FAQ (or the 20 most common support-inbox questions if your FAQ is smaller). Run each question in a fresh ChatGPT, Perplexity, and Gemini session. Check whether your store is cited, whether the answer text matches your FAQ answer, and whether your answer appears verbatim or paraphrased. That last signal — verbatim vs. paraphrased — is the cleanest indicator of whether your FAQ writing quality is at the extraction threshold.

Cited + verbatim

Ideal. Your answer is at the extraction threshold and retrievers trust it enough to quote directly.

Cited + paraphrased

Good. Your answer is in the pool but being reworded — often because the original is too promotional or too long to lift cleanly.

Not cited, competitor cited

Your content is in the pool but not winning the passage comparison. Check the competitor's answer and see which of the six rules you are missing.

Not cited, no one cited

The question is not producing citations on any engine — either the query class is not suited to FAQ answers, or the retrieval pool for this query is failing entirely.