If a language model can turn out a passable explainer on any topic in a minute, a small site can't win by publishing more explainers. Anyone can match that output for free, including the people competing with you for the same search results.
What a small site can still do is publish things a model had no way to write: what you tested, what you measured, what went wrong, and what you concluded from it. Put that under a name readers recognize, and build a way to reach those readers that doesn't depend on a search result.
That's the strategy. Below is what each part looks like in practice, what Google's own documentation says about AI content, and a checklist to run on any article before you publish it.
What AI made cheap
A model writes from text that already exists. So the content it replaces most easily is the kind that was already a rewrite of other pages: definitions, "what is X" summaries, generic how-to steps, product roundups assembled from spec sheets and other people's reviews.
If a page on your site could have been produced by someone who has never done the thing it describes, assume a model can produce it too. Those pages aren't worthless, but they're the ones with the least protection.
Five things a model can't supply for you
First-hand experience, with evidence. A model has never used the product, run the campaign or filled in the form. You have, or you can. The evidence is what makes it believable: your own screenshots and photos, the settings you used, the dates, the numbers, and the mistakes. A paragraph on what went wrong and what you'd do differently is often the most useful part of a page, and it can't be written by summarizing a sales page. If you write reviews, the outline in my guide to writing a buyer's guide is built around this kind of proof.
A specific point of view. Ask a model which option is best and you'll usually get a balanced list of considerations. Readers mostly want a recommendation and the reasoning behind it: "I'd pick this one unless you need that feature." Taking a position means you'll sometimes be wrong in public. I think that's a fair trade, because a page that commits to nothing gives nobody a reason to come back.
Original data or a tool. New information is the one thing a rewrite can't contain. It doesn't have to be a big study. A spreadsheet comparing the terms of ten programs you checked yourself, a small survey of your subscribers, a calculator, or a template people can copy all count. Say how you collected the data and when, so readers can judge it.
A direct relationship with readers. Search traffic is borrowed. An email list is the simplest thing you own: people who asked to hear from you and whose replies tell you what to write next. If you don't have one yet, start with a five-email welcome sequence rather than a complicated funnel. A comment section or small community you take part in does a similar job.
A name people trust. This is a real person on the byline, an about page that says who you are and how you make money, corrections when you get something wrong, and the same standards applied to every post. It builds slowly, and one invented claim can undo a lot of it.
What Google actually says about AI content
Google doesn't penalize a page just because AI helped write it. Its guidance on using generative AI content says the tools can be useful for research and for adding structure. It also warns that generating many pages without adding value for users may violate the spam policy on scaled content abuse, and it tells you to check AI output for accuracy before publishing.
The spam policies define scaled content abuse as "when many pages are generated for the primary purpose of manipulating search rankings and not helping users." The examples on that page include AI-generated pages, and also scraped or stitched-together content made by any method. The test is purpose and value. Which tool you used doesn't decide it.
"Experience" is the first E in E-E-A-T (experience, expertise, authoritativeness, trustworthiness). Google's page on creating helpful, reliable, people-first content includes this self-assessment question:
Does your content clearly demonstrate first-hand expertise and a depth of knowledge (for example, expertise that comes from having actually used a product or service, or visiting a place)?
Be careful with how E-E-A-T gets described elsewhere. The same page says "E-E-A-T itself isn't a specific ranking factor," and that Google's systems use a mix of factors to identify content that shows it. Trust is named as the most important of the four. So an author box or a "reviewed by" line doesn't switch anything on. The page content has to show the experience.
That page also suggests telling readers how a piece was produced when they'd reasonably wonder, including whether automation was used.
