AI content and FAQs on a business website after Google’s 2026 changes: what still makes sense

What exactly disappeared from search in May 2026

In the Google Search Central documentation changelog, under the date May 8, 2026, there is an entry about deprecating the FAQ rich result: the feature would no longer appear in search results starting May 7, 2026. A later changelog entry followed — the documentation for the FAQ rich result was removed, because the feature is no longer shown. The announcement concerned a form of presentation: the expandable list of questions and answers under a link in the results. It did not concern the content of your website.

Untangle three things that rumors merge into one: the page content (it stays exactly as you wrote it), the structured data (code that describes that content), and the rich result (the form of display in search — this is what disappeared). A client who lands on your site from maps, an ad, or a recommendation reads exactly what you wrote, regardless of what happens in the results. One display surface you never controlled has closed.

Why an answers section still makes sense

Losing the expansion under the link does not take away the answers section's daily job. A client opens your site in the evening, before calling, and wants to know three things: do you do what they need, how long does it take, and what should they prepare. If the answer stands on the page, the phone call starts with the actual matter instead of a repeat of the basics — and that is the only measure of this section that Google never withdrew.

Run a simple test on your own materials: take the last answer you sent by email and check whether it sounds like text from your website. If it does, it is ready to publish almost unchanged, because that is exactly how people ask. If your answers section is built from phrases nobody would say out loud in conversation, it is not a section for clients — it is writing for robots, and since May 2026 it does not even earn the expansion in the results. The same mechanisms that stop clients before they hit the send button are described in our analysis of why customers don't leave inquiries — they apply to your list of answers too.

That same list then works in several places at once: on the website, in your email reply templates, and in onboarding for new staff, who in their first week get a ready cheat sheet of your real topics.

How Google sees AI content: where the spam line is

Google's documentation on generative AI content says it plainly: you can use these tools for research, organizing a topic, and adding structure to original content, but generating many pages without adding value for users may violate the scaled content abuse spam policy. There is no ban on AI here — there is a condition of value. The same document reminds you to focus on accuracy, quality, and relevance when creating content automatically, and it lists specifics: title elements, meta description elements, structured data, and alternative text for images.

There is also a sentence from the changelog that puts 2026 in order: the spam policies apply to all of Google Search, including generative AI responses. A page that exists only to exist gains nothing — neither in classic results nor in assistants' answers.

The line is simpler than the policy names suggest. Run one test before publishing any page: "who will read this to the end and be glad they did?" If you have no answer, don't publish — no matter whether a human, a model, or both wrote the text. This criterion guards against both errors: an empty page mass-generated overnight and an artificial text hand-written for a keyword.

Where Google takes the description under your link from

Snippets are primarily created from page content; Google may use the meta description element when it describes the page better than content taken directly from the page. Two practical things follow. First, the description under the link is not the place for a slogan: write the meta description as one sentence for a human that says what the client will find on the page and who it is for. Second, since the snippet is built from content, a site without clear sentences about services, prices, and scope leaves that description to chance — Google will stitch together whatever it happens to find.

The documentation also lists the limiting tools: nosnippet, max-snippet, and data-nosnippet let you control what may be shown under the link. A small business rarely needs them — it is enough that the page content and its description say the same thing. The common mistake runs the other way: a description written around a superlative ("the best service in town") rarely describes the page better than its own content, so the search engine simply skips it.

When AI helps and when it hurts

A language model does mechanical things well when the source material is good, and does badly at things that require your knowledge of your own business. The table gathers practices from both sides of that line.

What's worth delegating to the model

A draft answer to a question that keeps coming back — the model will organize material you give it from a call transcript or an email thread, and you decide what stays. Also good: headline variants for a single page, shortening a paragraph that grew too long, and a list of services drawn from a list you supply. Always with one condition: the source is your material, not the model's "knowledge" of your industry.

What not to generate at scale

Separate pages for every city where clients might theoretically search; descriptions of services you don't perform; numbers and "studies" nobody at your company ever ran. An example on made-up numbers — plug in your own: ten cities means ten pages carrying the same answer, while one good page with ten answers gives the reader exactly as much, without the risk under the scaled content abuse policy.

Before you publish: facts, prices, terms, regulations

Every sentence on a company website that can be checked should have a path to its source. Verification doesn't need to be a corporate process — before publishing, walk the list and, at each fact, answer the question "how do I know this?"

Where to take numbers and regulations from

• prices and service scope — from your current price list,
• terms and hours — from the calendar, not from memory,
• regulations — from the text of the act or an official site (uodo.gov.pl, gov.pl), not from a podcast,
• numbers and quotes — from a named source you can point a client to,
• service names — from the list of what you actually deliver.

One person signs for the whole thing: you, or someone who knows the business from the inside. The model does not sign for content — you do, so the last word on facts belongs to a human. If a sentence starts to sound like legal advice, replace it with a description of your procedure and refer to a lawyer or a data protection officer; a company website describes, it does not adjudicate.

One sentence about how the text was made

Google suggests considering adding information about how your content was created — for example, brief background on how automation was used, in a form that makes sense for your audience. For a service business that is usually one sentence at the bottom of a page: "We collect answers from calls and emails, we draft them with an AI tool, and a human checks the text before publishing." That sentence needs no confession and no manifesto — it simply tells the reader what they are looking at.

For online stores the matter is formalized: Google Merchant Center requires that automatically generated product titles and descriptions be labeled as AI-generated, and that AI-generated images carry the appropriate IPTC metadata. If you run a store, this is not a choice but a platform requirement.

Do it yourself: ten answers from real conversations

A list works better than theory, so here is the concrete version — one evening of work:

1. Open your inbox and the last month of phone notes; write down every client question that repeated.
2. Pick the ten topics that come back most often — not the ones that sound "strategic".
3. Answer each topic in three to five sentences, the way you answer on the phone; the model may draft, but the source is your answer.
4. Check prices, terms, and hours against your own records, and regulations at the source.
5. Put it all on one page as a simple list; drop the idea of separate pages per city.
6. Return to the list after the season and fix whatever went stale.

If you are building the site from scratch, the same rule applies as in the Websites and stores service: the site stands around one decision the client has to make — order, book, or call — and the list of answers helps that decision instead of drowning it. More on the order of such work in a small business: where to start automation.

What it looks like when a system gathers the answers

The same process can run continuously instead of once a season. At Aura, the system collects questions from calls and emails, builds them into a list of repeating topics, prepares draft answers, and hands them to a human for the decision — and the approved answers land on the website and with the assistant that uses them in conversations with clients.

The human keeps two things: approving the facts (prices, terms, scope) and deciding what gets published at all. The rest — collecting, sorting, reminding about refreshes — is the system's work. The risk is the same as with manual AI writing: a model can sound confident while being wrong, which is why the decision never passes to it. No promises of positions in Google or citations in assistants' answers — what you get is material that both the search engine and your clients can use.

Texts in the context of the whole site and real client questions — headlines, the offer, service descriptions, and calls to action — are written by our Copywriting service. A consistent description of your business for search engines and assistants, without citation promises, is kept in order by GEO / AI visibility, and visibility in Google and maps within a few streets, with a profitability threshold formula, is handled by SEO and maps. After hours, the same answers are served by the AI Chatbot — a site assistant that replies from your knowledge base and introduces itself as AI from its first sentence.

We write about the difference between a standalone agent and a chatbot separately: AI agent vs chatbot. For restaurant owners we also recommend our pieces on the facts assistants read about you and on AI in guest service and what Polish law says.

Frequently asked questions

Is the answers section on a website pointless after May 2026?

No. What stopped being shown on May 7, 2026 is the FAQ rich result — the form of presentation in search. The answers section keeps working on the same terms: clients read it before contacting you, and you get fewer repetitive conversations. The criterion is unchanged: the answer has to help the person looking for it.

Do the FAQ structured data markup have to be removed?

There is no such requirement. Google announced the end of displaying that result and removed its documentation, but the markup is an element of your site's code — that decision is yours together with your technical person. If it was added solely for the expansion in results, that reward no longer exists.

Does Google penalize sites for AI-written content?

There is no ban on AI content. What breaks the policy is different: generating many pages without value for users, that is, scaled content abuse. Using tools for research, structure, and drafts is allowed — while watching accuracy, quality, and relevance, including title elements, meta descriptions, structured data, and alternative text.

Does the meta description still matter?

Yes, but not as a spell for rankings. Google builds the snippet under the link primarily from page content and may use the meta description when it describes the page better than content taken from it. Writing the description as one sentence for a human increases the chance that it is the one shown.

How do I show clients that a text was written with AI?

Google suggests considering adding information about how the content was created — for example, one sentence about what automation was used for. For stores the rule is stricter: Merchant Center requires automatically generated product titles and descriptions to be labeled as AI-generated, and AI images to carry the right metadata.

How many answers should one page hold?

There is no official number. The practical limit is the volume a client will actually read: ten topics that genuinely come back in conversations make sense on one list, while an artificially inflated encyclopedia makes sense nowhere. What counts is whether each answer sounds like something you would say on the phone.