Google's content rules, minus the legalese

Nobody actually reads Google’s guidelines. They quote a tweet about them, or read a LinkedIn thread written by someone who skimmed a tweet about them. Meanwhile the documents sit there, updated every few months, quietly deciding which pages rank and which ones get a manual action.

I read the whole thing recently — the spam policies and the helpful content guidance. It is shorter than you’d expect, and it says more useful things than any of the secondhand versions floating around. Here is the version I wish I’d found first.

The one rule everything hangs on

Google has exactly one sentence doing most of the work: content is for people first, not for search engines. That sounds like a fortune cookie until you look at how they operationalize it. They published a self-assessment checklist, and it is basically four questions in a trench coat:

• Do you have a real audience that would want this if it never touched Google?
• Do you actually know the thing, from doing it — using the product, visiting the place, running the process?
• Does the reader leave knowing more than when they arrived, without needing to search again?
• Is this your site’s job? A focused site beats a blog about everything.

Everything else — E-E-A-T, the quality rater guidelines, “Who, How, and Why” — is just this idea with paperwork attached.

One detail worth repeating, because half the SEO industry still gets it wrong: Google has no preferred word count. They say so explicitly. Writing to 1,800 words because a tool told you to is optimizing for a fiction. Same for the “keep content fresh” trick of nudging dates without changing the substance — they wrote a whole paragraph telling you it does not work.

Trust is the actual metric

E-E-A-T — experience, expertise, authoritativeness, trustworthiness — is not a checklist you can add as schema and call it a day. Google says trust is the one that matters, and the others only feed into it. Their shortcut for measuring it: “Who wrote this, how was it made, and why does it exist?”

Who: a byline that links to a real person, on an author page that links back to things they’ve actually done. How: for reviews, evidence you tested the product; for AI-assisted drafts, an honest disclosure of where the automation ended. Why: the honest answer should be “because our audience needed this.” If the honest answer is “because this keyword had volume,” that is exactly what the helpful content systems are built to find.

For anything touching health, money, or safety — the YMYL zone — the bar goes way up. Google openly weights these topics harder.

What Google now says about AI content

AI-written content is not banned and never was. Google’s position since 2023 has been: whoever or whatever drafts it, you own the outcome, and useful is the only bar that matters.

The trap is a different policy: scaled content abuse. Since March 2024, it doesn’t matter how the content was made — AI, scraped feeds, stitched pages, synonym-spinners. Making lots of pages with little original value, at scale, to game rankings, is spam. The volume was always a fingerprint of the intent.

The 2026 update that almost everyone missed: in May, Google clarified that these spam policies now explicitly cover generative AI responses in Search — AI Overviews and AI Mode. The rules stopped being just about blue links. Manipulating what an AI answer says counts the same as manipulating a ranking.

Two practical notes from this year’s docs:

• Disclose AI use where a reasonable reader would wonder. “How was this made” is a trust question, and dodging it is itself a signal.
• If your workflow runs through AI drafts, give them a real human editing pass before they ship. We keep a small open-source tool for exactly this — OmniScribe — but the tool is optional. The editing pass is not.

The don’t list, in plain language

The spam policies read like a list of ways to lose your site. Quick translation:

• Cloaking — showing Google a different page than your visitors get. Paywalls are fine if Google sees the same content a subscriber does.
• Doorways — thin pages that only exist to funnel searchers somewhere more useful.
• Expired domain abuse — buying someone’s old domain and renting out its history to your junk.
• Hidden text and links — white-on-white, zero opacity, text shoved off-screen. Accordions and tabs are fine; the policy targets things invisible only to humans.
• Keyword stuffing — the city-and-phone-number footer, the paragraph where the brand name appears every third word.
• Link spam — buying links, swapping links, guest-post farms with keyword-stuffed anchors, footer link networks. Paid links aren’t banned; they just need rel="sponsored" or rel="nofollow".
• Machine-generated traffic — automated rank-checkers hammering Google. Use an official API.
• Malicious practices — malware, unwanted software, and, new as of April 2026, back button hijacking — the trick where hitting back on a page sneaks the visitor somewhere else. That’s now its own named policy.
• Misleading functionality — fake converters and tools that exist to bounce you through ad labyrinths.
• Scraping — republishing other sites’ content without adding anything of your own.
• Site reputation abuse — “parasite SEO.” Renting your domain’s authority to third parties. A coupon site running an anonymous “best casinos” page it didn’t write is the canonical example. As of August 2026, inside the EEA the penalty changed: those pages get ranked separately from the host domain instead of hammered manually. Outside the EEA, the manual actions continue.

One thing that stung to learn: since April 2026, spam reports people file can feed manual actions. Competitors are, in effect, part of the enforcement system now.

Quiet deaths worth knowing about

Google spent 2025–2026 trimming the search results page, and some SEO habits died with features that no longer exist:

• FAQ rich results are gone — removed May 7, 2026. Shipping FAQ schema for the ranking benefit is effort spent on nothing. Keep it only if it serves readers on-page.
• llms.txt does nothing for Google — confirmed June 2026. Keep it for other systems if you like; just don’t expect Search to care.
• Course info, practice problems, special announcements, estimated salary, vehicle listings, and a few others got retired. Check your schema audit against the current list before the next sprint.
• Breadcrumbs now show only on desktop.
• On the up side: preferred sources — the feature where users can pick which outlets they see — rolled out to all languages and now surfaces inside AI Overviews and AI Mode. If AI answers are part of your strategy, that’s a lever worth pulling.

What this means if you’re doing AEO

Here’s the part I find quietly satisfying: Google’s people-first standard and the standards that make you citable in AI answers are converging on the same thing.

An answer engine cites you when a passage is clear, specific, and defensible — first-hand detail, named authors, sources. That is the same checklist as “Who, How, Why.” The site trying to win ChatGPT citations with a wall of AI-summarized commodity content is the same site Google’s scaled content abuse policy was written to catch. The two games reward the same behavior, which makes the strategy a lot simpler than most takes suggest: make the best page on the internet for the thing you actually know, show your work, and let the humans (and the models) verify it.

The people who lose in both systems are the people optimizing for neither.

If you want to see whether AI engines actually name your brand today — before optimizing anything — start with measurement. Our free AI visibility check takes a prompt and shows you who gets cited, and it takes about a minute. For the broader strategy, the AEO vs SEO comparison covers how the two games differ and where they overlap.