FAQ on your site for Alice and AI assistants
Alice answers from fragments it understands about your market. The best commercial fragment is a normal FAQ: the question the way a client asks it, the answer with a price, a timeline, and a place. Not "we are market leaders", but "how much does an iPhone 14 screen replacement cost in Kazan, from 4,500 ₽, two hours".
Below: where to get the questions, how to put them on the site, what kills the chance of a recommendation, and how to check the result out loud and in an audit.
Where the questions come from
Calls, WhatsApp, Avito, reviews that contain questions, on-site search. Fifty real phrases beat a hundred invented "for SEO". Write them the way people speak, not the way a copywriter would.
- How much does … cost
- How fast …
- Where are you …
- Is there a warranty …
- Are you open on weekends …
One block on the homepage will not save you if you have many services. You need an FAQ on the service page plus a general one on contacts or "about". On a query about implants, Alice looks at the implant page, not the slogan on the homepage.
How this FAQ differs from an SEO block
An SEO block says "we are leaders", "a personal approach", keywords in the headings. An FAQ for AI assistants says "how much", "how fast", "where are you", with numbers in the first sentence.
Clients rarely read the FAQ. Alice and Yandex Neuro read it whenever they assemble an answer. Empty marketing means an empty recommendation.
Ten questions with prices beat a hundred questions with no numbers. Volume does not win. Specifics that match the maps do.
How to format it so people and bots can both read it
The answer belongs in HTML, not in a PDF and not in a picture of the price list. An accordion is fine if the text is in the DOM and is not loaded empty. Two to four sentences with numbers on each item. A link to booking or to the price list. The phone at the end is the same one as in the footer and in 2GIS.
Add FAQPage schema with the same questions and answers, word for word. A gap between the schema and the visible text costs trust. More: Schema.org.
The first sentence of the answer is the direct answer, with a number or an address. "We aim for quality" will not land in a voice answer. "Implants from 45,000 ₽, consultation free" will, if it matches the maps.
What not to write
General words with no prices. "A personal approach" with no timelines. Questions nobody asks. Ten pages of the same text copied across cities. Alice and Yandex Neuro look for specifics, not volume.
Examples by niche. Dentistry: implant price, does it hurt, installment plan, private insurance (DMS). Auto repair: repair time, parts, warranty. Delivery: minimum order, zone, time. Take them from the CRM, not from a keyword generator.
Five signs the FAQ is not working
- A copywriter invented the questions. Support calls did not.
- Answers with no numbers: "a personal approach", "high quality".
- The FAQ is only on the homepage. Service pages are empty for AI assistants.
- The accordion text is not in the DOM. The bot sees nothing.
- FAQPage schema does not match the HTML word for word.
Fail one item and Alice has nothing to build an answer about you from. The competitor with a table of questions takes the recommendation.
How an FAQ gets into a voice answer
Alice does not read a site the way a person does. It assembles a short fragment: question plus answer with a number. A long marketing paragraph does not fit in a voice reply. The model takes the first sentence that contains a fact, or it goes to a competitor.
The query "how much does an oil change cost nearby" looks for a service page with an FAQ or a price list. A homepage that says "we care about your car" is not a candidate. A narrow query needs a narrow page.
Yandex Neuro in the app works the same way: one connected answer. An FAQ on the implant page beats an FAQ on the homepage for a query about implants. That is not SEO cannibalization. It is precision for an AI assistant.
ChatGPT more often cites what GPTBot read in the HTML. A PDF price list inside the FAQ will not land in the answer. An HTML table with three rows will, if robots.txt is open.
After you publish the FAQ, check: Ctrl+U, the text is in the DOM; FAQPage schema matches word for word; three queries spoken out loud. If you are not in the answer, compare the first paragraph with the competitor from the audit. The difference is often one number.
Typical case: a clinic wrote an FAQ that said "a personal approach for every patient". On "how much does an implant cost", Alice names the neighbor with "from 45,000 ₽, consultation free". That is not style. It is a fact in the first sentence.
Update the FAQ when prices change. Last year's number in the HTML next to a current number on the maps is a conflict. The model picks the competitor whose facts match.
15 to 25 questions from the CRM are enough to start. A hundred questions with no numbers are worse than fifteen with prices. Volume does not win in AIO.
What people do themselves, and where it breaks
An FAQ "for SEO"
Ten marketing paragraphs under question headings. The AI assistant does not quote them. The client does not read them. The time is spent.
A PDF instead of text
The price list is a file. The FAQ is a picture. The bot sees a link and does not see the numbers. Alice names the business whose prices are in HTML.
One FAQ for the whole site
A global block with someone else's questions from a template. On a narrow query, emptiness. You need questions on the service page.
Schema with no HTML
FAQPage in JSON-LD, a different text on the page. Trust drops across the domain.
Alice and Yandex Neuro
After you publish, check three voice queries for the niche: "who would you recommend", "how much does it cost", "who is closer". If you do not hear yourself, compare the FAQ with the competitor from the audit. They often close the same questions with numbers. You closed them with marketing.
Voice scenarios: Alice recommendations. Weak Yandex Neuro and weak Alice often travel together. They are one layer of facts on the site and on the maps.
Order of work
- Audit: is there an FAQ, and can bots see it.
- 15 to 25 questions from the CRM and from calls.
- Markup plus FAQPage schema, word for word.
- Match it to the maps and the price list.
- Audit again, and test in Alice.
Where to start today
Do not write an FAQ from a copywriter's head. Do not copy someone else's block from another site. Start with a snapshot in the audit. You will see whether what is already there can be read.
Without an FAQ you leave the AI assistant an empty box. A competitor with a table of questions takes the recommendation, even if your design is nicer. Pricing: LeadsyLabs.
Every week without a proper FAQ is Alice answers with other names. That is measurable: three queries out loud, three times in a row other brands, and you are not in the recommendations.
A voice scenario: "how much does [service] cost nearby". Alice does not read PDFs. It takes a short answer from the FAQ or the listing. No answer, no brand in the recommendation.
After the edits, do not wait for an instant miracle. Bots have to read the new text first. Then the model updates the picture. A week is a normal minimum before you test again.
"We have an FAQ" and "Alice quotes us" are different things. A block of marketing will not land in a voice answer.
A short checklist: 15 or more questions from the CRM, numbers in every answer, schema word for word, three spoken tests, an audit again in a week.
A free audit shows whether bots can see your FAQ. Without that, every new paragraph is a lottery.
Who writes the FAQ: not a copywriter from nowhere. Sales, support, the owner, the people who hear the questions. A developer marks it up and adds the schema.
Reviews inside the FAQ do not replace reviews on the maps. The question "who has good reviews" is closed with a link to 2GIS and an honest number, not testimonials with fake faces.
If the FAQ is green after the audit, still run three queries in Alice. The number in the report and the voice from the speaker complete each other.
Questions on implementation: FAQ. A full audit with an action plan for the FAQ and schema: pricing.
Typical case: a copywriter wrote the FAQ over a weekend and nobody checked the prices against the maps. On "how much does it cost", Alice names the competitor, because that competitor's numbers match. The fix is not style. It is support calls, and schema that matches word for word.
A voice scenario behind the wheel: "where is [service] nearby". Alice does not read a PDF. It takes a short answer from the FAQ or the listing. No answer, no brand. Three seconds, and the client is with a competitor.
Accordions on Tilda and other builders: check Ctrl+U, not only the preview. The text has to be in the HTML. It must not load empty on click.
On a multi-page site: a general FAQ on contacts, plus a block on each service. One global accordion will not cover "implants in [city]". You need an implant page with questions about implants.
Prices in the FAQ and on the service page have to match the maps. A 500 ₽ gap, and the model picks the business whose numbers match everywhere.
Who owns the FAQ in the company: the owner approves prices, support supplies the wording, the developer adds schema. A copywriter with no CRM access is the usual failure.
Tied to AIO and GEO Score: after robots.txt and NAP, the FAQ is one of the fastest fixes.