Buying guides and FAQs that AI assistants can quote

A buying guide that lists "five things to look for" and stops there does not answer the question the index's prompts ask. We know the form of that question because we wrote it. Dopbot's index asks a buying-advice question in each of the 40 US categories it updates monthly from November 2026, and each category page shows its prompt once the November update is published. This post uses those prompts as the specification for a guide, then shows what our October 2026 data can and cannot say about advice-type answers.

What do buying-advice questions look like?

In the November 2026 edition, the advice question is the fifth of five prompts per category. It names no brand and no retailer. It is described in the methodology, and its answers start with the November 2026 run, so the October 2026 edition holds no answers to it. This post uses only the wording and our October data; the advice answers are on the category pages once that run is published. Six of the 40 prompts, as written:

Reading all 40 prompts side by side, the pattern is regular:

• All 40 are two sentences: first a statement of the shopper's situation, then a question. The mean length is 21 words.
• All 40 ask "which" products. 30 of the 40 contain "would you recommend". Only 1 of the 40 also asks what to look for, in a carry-on suitcase question from a frequent flyer.
• The situation carries the criterion. None of the 40 prompts says "firmness"; the mattress prompt says the shopper's back aches. A guide has to translate the symptom into the spec.

The last point is the practical one. A question that asks "which would you recommend" is best answered with products, so criteria alone leave half of it open. A guide that says "look for medium firmness" and never names a product hands an assistant half of what the question wants. The other half has to come from somewhere else, and that somewhere may be a competitor's page.

What do the October answers say about advice-type questions?

Not much directly, but two things are worth reading. The first is how often an answer recommended no brand, by prompt type. The October 2026 edition used five prompts, none of them a buying-advice question. The closest thing we can measure is how often an answer recommended no brand at all; we did not read those answers, so we cannot say what they said instead.

October 2026 edition, OpenAI's API with web search (gpt-6-luna) and Perplexity's Agent API with the Sonar model (perplexity/sonar) combined in each row. These are recommended-brand counts, not source counts, so pooling the engines here does not mix source lists.

The relationship is descriptive. We did not test why the open question more often ended with no recommended brand, and 9.1% of 318 answers is about 29 answers.

The second is which health, sleep and pet-care sites OpenAI linked. Among the 30 domains OpenAI's answers linked most often, four are sites that publish health, sleep or pet-care information: sleepfoundation.org, a sleep advice and mattress review site; aad.org, the American Academy of Dermatology; wsava.org, the World Small Animal Veterinary Association; and akc.org, the American Kennel Club. Each engine is counted on its own answers, here OpenAI's 797 answers with at least one link.

Pages linked in OpenAI's API answers (gpt-6-luna), October 2026 edition, 797 answers with at least one link; categories are out of 160. None of the four is among Perplexity's 30 most frequent retrieved domains.

We did not read these pages for this post, and we cannot say they were the advice sections of those sites. A domain in a source list does not show that it shaped an answer. The only safe reading is narrower: in some answers, in 5 to 9 categories each, OpenAI linked a page on a site that publishes sleep, skin, veterinary or dog-care information. We do not know which prompts produced the links, or whether the pages were guidance or product reviews.

How should a guide be built?

Some facts are documented, and they are fewer than the usual advice implies. As of October 2026, Google says there is no special optimization for its AI features beyond ordinary search practice, that a page must be indexed and snippet-eligible to be a supporting link, and that no special structured data is needed. OpenAI says OAI-SearchBot is the crawler that surfaces sites in ChatGPT's search features. Perplexity says PerplexityBot is built to surface and link sites in its search results. None of the three documents explains how a passage is scored or chosen to be quoted.

So the structure below is our recommendation, built from the prompts above and from Google's published guidance on helpful content. Treat it as good practice to test, not a recipe with a known payoff.

1. Start from real questions. Collect them from support tickets, return reasons, review text and the prompts above. Write each as a situation, the way the shopper wrote it.
2. One situation per section. Give each section a heading in the shopper's own words, such as "Which pillow for a stomach sleeper with a stiff neck?" A heading that matches the question helps a reader find the passage without the rest of the page, and it may help a retrieval system too; no source we found says so.
3. Open every section with a direct answer. Two or three sentences: the criterion, the threshold that matters (with its unit and the reason it matters), and a named product that meets it. Details follow below it.
4. Name products, including when yours is not the right one. A guide that says when your product is the wrong choice is more useful to a reader and easier to trust. Name a competitor where it fits, and for each product say what it is for and what it is not for.
5. Show first-hand experience. Google's guidance asks whether content shows first-hand expertise, for example from having used a product, and whether it is written or reviewed by someone who knows the topic. Put the author, the testing and the review date on the page.
6. Keep it as text. Google lists important content available in textual form among its practices. A size chart pasted as an image, or an answer that only appears after a click or a script runs, may not be read by systems that work from the page's text, so put it in the HTML.
7. Add the product facts the passage depends on. Dimensions, materials, compatibility and return terms belong on the product page; the guide should agree with them exactly. The product pages audit covers that side.

A passage template, with placeholders:

If you [situation in the shopper's words], look for [criterion] of [threshold and unit], because [one-sentence reason]. [Product name] meets this with [specific measured attribute]. If [other situation], [other product or type] is the better fit, because [reason].

The aim is a passage that makes sense if lifted out of the page with nothing around it. That is a property of good writing for readers too, which is why it is a safe bet, though not a proven one.

Do FAQ sections and FAQ markup help?

A FAQ section helps readers find short answers, and the questions can be the situations above. FAQ markup is a separate matter. Google announced in August 2023 that FAQ rich results would appear only for authoritative government and health sites, and its changelog says the feature stopped appearing in Google Search on May 7, 2026. As of October 2026, FAQ markup will not produce a rich result for a store. If you add it anyway, Google's AI features guidance says structured data should match the visible text of the page. See structured data for AI shopping for the rest.

How do you check whether a guide is being used?

Run the test before and after publishing, on your own questions.

1. Write five advice prompts for your category in the shape of the table above, one situation each. Keep them fixed.
2. Ask each prompt in each assistant several times, in clean sessions, and record whether your brand is named and which source URLs appear.
3. Compute your share of answers per prompt with an interval. Three of 10 answers is 30%, but the 95% Wilson interval runs from 10.8% to 60.3%, so a change from 3 to 5 of 10 proves nothing. The share of answer post shows the method, and how to measure AI visibility gives the manual routine.
4. Change one thing, then repeat. Assistants also change between runs and between models, which is another reason to keep the prompts and the sampling identical.

What stays unknown after that is the app. Our samples come from developer APIs, and the apps can differ in personalization, location and product interface. A guide that raises your share in the API may behave differently in an app.

Frequently asked questions

How long should a buying guide be?

As long as the question needs. Google's guidance on helpful content lists writing to a word count as a warning sign and states that it has no preferred word count. Measure a guide by whether each section answers its situation, not by length.

Do the assistants show the same sources?

We did not measure page-by-page overlap. In the October 2026 edition, OpenAI's answers linked a mean of 2.2 pages each, and Perplexity's search retrieved a mean of 27.3 URLs per answer, a list that includes pages the answer may not use. The two lists are recorded differently, so compare each engine against itself. The websites in AI shopping answers post lays out each engine's most frequent domains.