How help centre articles get found by Google and AI assistants

Before a customer writes to you, they have most likely already googled the question or asked an AI assistant. If the answer that comes up is from your help centre, both the customer and you are spared a ticket. If it comes from a forum, a competitor or nowhere at all, you get the ticket, often with irritation thrown in. This article is about how a help centre is written and built so that it gets found, by Google and by AI assistants. There are no tricks in it; everything here helps the reader first.

Why is a public help centre searchable knowledge?

A public help centre is the only part of your customer service that can answer before the question is asked. The inbox and the chat help those who have already contacted you; the help centre helps those still searching. Every article is a page Google can index and an AI assistant can retrieve and cite, and every such hit is a ticket that never gets created.

That presumes the help centre is actually public, meaning it sits at an address that can be indexed and not behind a login or in a widget that only loads with scripts. Google writes in its documentation on AI features in Search that a page can appear as a source in AI Overviews and AI Mode if it is indexed and eligible to show a snippet in ordinary search, and that there are no additional technical requirements. The foundation is therefore the same as for all searchability: the page must be reachable, readable and understandable. A reviewed knowledge base thereby becomes your most-read channel, read by people who never become tickets.

How should a help article be written to show up in search?

One question per article, the question as the heading, and the answer in the first paragraph. That is what makes an article useful to a stressed customer, and the same properties are what make it easy for search engines and AI assistants to match against a question and lift out. The full template for such an article is in the article on writing help articles.

Google's guidelines on helpful, reliable content ask questions that are directly useful for a help centre: would someone arriving directly at your page find the content useful, does it show real knowledge of the product, and does the reader leave with their question answered? Google also describes that its systems weigh experience, expertise, authoritativeness and trustworthiness, with trust the most important. For a help centre that means the company itself answering, correctly and up to date, about its own products and terms.

Concretely:

• The heading is the question the customer asks. "How do I return an item?" instead of "Returns". Use the customer's words, not your internal ones.
• The first paragraph is the answer. Two to four sentences, enough for someone in a hurry. Details, exceptions and steps come after.
• One topic per article. Return and exchange are two articles linking to each other, not one long one.
• Steps as a numbered list, choices as a table. It can be skimmed, and it can be lifted out.
• Spell out what is specific to you. Dates, days, amounts, shipping methods. Nobody searches for generic wording.
• Define terms the first time. "Right of withdrawal" and "complaint" are not the same thing.

The hardest part is knowing which questions are missing. Go to your tickets and look for the questions you answer in the inbox without an article to link to; our article on finding the questions you cannot answer describes how.

What structured data does the help centre need?

Article markup on every article, and FAQ markup where the page really is a list of questions and answers. Structured data is machine-readable information in the page's code that says what the content is: an article, its heading, author and date, or a set of questions with answers. Google's documentation for Article describes that the markup helps Google understand the page and show a better heading, image and date in the result, and that it is open to all kinds of pages.

For FAQPage there are two things to know. The markup is intended for pages where the company itself asks and answers the questions, not for forums where users answer. And since 2023 Google only shows the rich FAQ results in search for well-known government and health sites, according to Google's announcement of the change. The markup is still correct to use and helps machines understand the structure, but do not expect it to give you a visible box in the results.

Most important of all: Google says explicitly in its guide to generative AI features that no special structured data is required to appear in AI Overviews. Mark up what is true about the page, and nothing more.

What is needed technically: sitemap, addresses and update dates?

A sitemap listing every article with a correct update date, one address per article that describes the content, and a visible date on the page. A sitemap is a file that lists the site's pages so search engines find them without following every link. Google's documentation on building and submitting a sitemap states that only canonical addresses should be included, meaning the address you want to count when the same content can be reached in several ways, and that the last-modified field should reflect the last significant update of the content, not every small change.

Addresses should be readable and stable: /help/return-an-item says more than /article/48213, to people and to machines alike, and an article that changes address loses the links and history it has gathered. Google's recommendations on URL structure apply equally to help centres: one address per piece of content, consistent spelling, no temporary parameters. What the accessibility act requires of the help centre is covered in the article on the accessibility act and customer service.

The update date is underrated. A customer searching "Christmas return rules" wants to see that the article applies this year. Show the date of the last substantive update on the page, use the same date in structured data and the sitemap, and change it only when the content changes.

How do AI assistants pick answers from the help centre?

AI assistants retrieve short, self-contained passages that answer the question directly and can be pointed back to with a source. ChatGPT, Perplexity and Google's AI Overviews differ in the details, but all of them search, read a number of pages and assemble an answer with references. Google describes in its post on how content performs in Search's AI features that both AI Overviews and AI Mode can run several related searches for a single question, and that the advice to publishers is the same as before: create unique, useful content and keep the technical foundation in order.

There is research on this too, under the name generative engine optimization. A study from Princeton among others, GEO: Generative Engine Optimization, tested how different changes to a source page affected how often and how much it was cited in AI-generated answers. Visibility could rise by up to 40 percent, and what helped was what makes the text more credible and concrete: citations, quotes and figures. Keyword stuffing did not help.

For the help centre that means in practice:

1. Every article should be quotable as is: heading, an answer in a few sentences, then details.
2. Spell out the company name and what is specific in the answer, so the answer makes sense even when lifted out of context.
3. State the date and, where there is one, which term or policy the answer is based on.
4. Keep the articles separate; an assistant cites the page that best matches the question, not the longest.

The same text that gets found by external assistants is the one your own chat should answer from. When help centre, chat and inbox share one knowledge base, as in the platform, the answer is right everywhere.

What to do

This can be done this week with the help centre you have.

1. Search Google for your ten most common customer questions, phrased the way the customer writes them. Note where your help centre lands, and what shows instead.
2. Ask the same questions to an AI assistant and see whether your help centre is cited. If not, read what is cited and compare it with your article.
3. Rewrite the headings of the ten articles as questions in the customer's words, and move the answer to the first paragraph.
4. Check that the help centre is in your sitemap, that the addresses are readable and that every article shows an update date.
5. Check that the articles have correct article markup, and FAQ markup only where the page really is a question-and-answer list.
6. Decide who owns each article and when it is next due for review.

Common questions

Does it help to put lots of keywords in help articles?

No. Google's guidelines on helpful content warn against text written for search engines rather than people, and the study on generative engine optimization found that keyword stuffing did not increase visibility in AI answers. What works is using the customer's own words in the heading and answering concretely. A search term the customer actually uses belongs in the heading once, not ten times in the text.

Do we need special structured data to appear in AI Overviews?

No. Google writes in its documentation that there are no additional technical requirements beyond the page being indexed and eligible for a snippet, and that no special schema.org markup is required for the AI features. Article markup and correct FAQ markup are still good to have because they help machines understand the page, but they should describe what is actually on the page.

How often should help articles be updated?

When the content changes, and otherwise at a fixed review, for example every quarter or ahead of seasons such as Christmas. The update date should only change with substantive changes; a date that changes automatically loses its meaning for customers and search engines alike. Give every article an owner who gets a reminder when it is time to read it through.

Can an AI assistant cite us incorrectly?

Yes. Assistants summarise and can mix up sources or use an old version of an article. What you can influence is that every article is unambiguous, current and self-contained, so it is hard to misread even out of context. Spell out the company name and the date in the answer, keep one article per question, and redirect old articles instead of leaving them up.

Sources

• AI Features and Your Website — Google Search Central, read 2026
• Creating Helpful, Reliable, People-First Content — Google Search Central, read 2026
• Learn About Article Schema Markup — Google Search Central, read 2026
• Mark Up FAQs with Structured Data — Google Search Central, read 2026
• Changes to HowTo and FAQ rich results — Google Search Central Blog, 2023
• Google's Guide to Optimizing for Generative AI Features on Google Search — Google Search Central, read 2026
• Build and Submit a Sitemap — Google Search Central, read 2026
• Ecommerce URL Structure Best Practices — Google Search Central, read 2026
• Top ways to ensure your content performs well in Google's AI experiences on Search — Google Search Central Blog, 2025
• GEO: Generative Engine Optimization — Aggarwal et al., Princeton University et al., arXiv, 2023