Google's own content guidance contains more direct denials than most vendors' advice contains claims. No preferred word count. No ideal keyword density. No required heading structure. No special formatting to appear in an AI Overview. Most "SEO content writing" advice sells a formula Google has explicitly said it does not have, and the vendors selling it rarely link to the page where Google says so, because the page undercuts the pitch.
That gap matters more now than it did before generative answers existed, because there is genuine, checkable evidence about what does improve a page's chances, both of ranking and of being cited in an AI-generated answer, and it comes from two very different sources that happen to agree with each other: Google's own stated content-quality guidance, and a peer-reviewed academic study specifically measuring what gets cited. Neither supports a word-count target, a keyword-density formula or a schema trick.
What Google actually asks for
Google's core statement is a definition, not a checklist: "Google's automated ranking systems are designed to prioritize helpful, reliable information that's created to benefit people, and not content that's created to manipulate search engine rankings." Its self-assessment questions operationalise that without producing a formula: does the content provide original information, reporting, research or analysis, does it offer a complete description of the topic rather than a shallow one, would a reader want to bookmark or recommend it, is it the kind of page mass-produced across a large network of sites rather than given individual care.
The expertise questions go further than most vendor checklists bother to: does the content show "clear sourcing, evidence of the expertise involved, background about the author," and is it "written or reviewed by an expert or enthusiast who demonstrably knows the topic well." Google draws the line between SEO and content quality explicitly rather than treating them as opposites: "SEO can be a helpful activity when it is applied to people-first content, rather than search engine-first content." The warning signs for the latter are named directly too: writing mainly to attract search visits, using automation to produce content across many topics, summarising what others say without adding value, and backdating pages to look freshly updated when nothing substantial changed.
The specific claims Google rules out
Word count. Google's own self-check asks whether a writer is "writing to a particular word count because you've heard or read that Google has a preferred word count," and answers itself: "No, we don't." Any guide recommending a specific target length as an SEO requirement is stating something Google has directly denied.
Keyword density. There is no Google-confirmed ideal percentage. Vendor rules of thumb exist (commonly cited figures sit around 0.5 to 3%), but they are vendor convention, not Google guidance. A Google spokesperson has been quoted more usefully than any percentage: mentioning a term once or twice can help ranking, "but just because you say it seven or eight times, that doesn't mean it will necessarily help your rankings." Google's SEO Starter Guide is blunter still about the failure mode: "keyword stuffing is against Google's spam policies," and its title-link guidance gives the example of what not to do, "Foobar, foo bar, foobars, foo bars," as a pattern to avoid rather than imitate.
Heading structure. There is no required hierarchy: "From Google Search's perspective, it doesn't matter if you're using them out of order," and no ideal count either, "there's also no magical, ideal amount of headings a given page should have." Headings are for the reader's benefit, not a ranking mechanism with a correct shape.
Featured snippets and AI Overviews. Google states there is no exact minimum length required to appear as a featured snippet, and no additional structural requirement to appear in AI Overviews or AI Mode: "There are no additional requirements... nor other special optimizations necessary." Content written to a supposed "AI Overview format" is being optimised for a specification Google says does not exist.
Readability evidence that is real, and correctly labelled
Structure genuinely affects whether content gets read, which is different from claiming it affects ranking or citation directly, and the two get conflated constantly. Nielsen Norman Group's original 1997 usability testing found 79% of test users scanned a new web page rather than reading it word for word, with only 16% reading in full; scannable formatting, sub-headings, bulleted lists, one idea per paragraph, measurably improved usability in the same study. Their later eye-tracking research describes an "F-shaped" reading pattern: a full horizontal scan near the top, a shorter horizontal scan further down, then a vertical scan down the left-hand side, which is why front-loading the key point in the first two paragraphs and starting headings with information-bearing words is genuinely evidence-backed, as a readability practice. None of this is Google-confirmed ranking research, and none of it is AI-citation research either; it is independent usability testing, and it is worth being honest about which category it falls into rather than presenting it as an SEO or GEO finding it was never designed to measure.
Two schema types that stopped working, still recommended everywhere
FAQPage structured data was restricted to "well-known, authoritative government and health websites" in September 2023, then removed from Google Search entirely by mid-2026: "The FAQ rich result feature is no longer shown in Google Search results." HowTo structured data was removed from search results the same week in September 2023. Both remain standard recommendations in current "SEO content writing" advice years after Google stopped showing either rich result. Marking up a page with schema that once earned a visual result and no longer does is not harmful, but it is not the visibility tactic it is still sold as.
What genuinely does move the needle, with real numbers
The clearest evidence on this comes from the academic paper that coined the term GEO (Generative Engine Optimization), tested across roughly 10,000 queries in its own benchmark. Measured against an unoptimised baseline, adding quotations improved visibility in generated answers by 41%, adding statistics by 33%, citing sources by 28%, and improving the fluency of writing by 29%. Keyword stuffing did the opposite of what SEO folklore assumes: it performed worse than the unoptimised baseline. The same paper validated its findings on a live, deployed generative search engine, not only its own offline benchmark, where statistics addition improved a separate "subjective impression" measure by 37%.
Google's own content-quality checklist points at the same underlying behaviour without citing a study to prove it: original information and analysis, genuine sourcing, demonstrable expertise. The academic evidence and Google's stated guidance agree, which is itself worth noting given how rarely vendor "GEO tactics" content bothers to check either.
What organic rank does not guarantee
Worth knowing before assuming a page that already ranks well is automatically covered: an analysis of 200,000 Google AI Overviews found a page ranking first organically appeared inside the AI Overview only 46% of the time on desktop and 34% on mobile, and only 20 to 26% of AI Overview citations overlapped with the top 10 organic results at all. Ranking well and being cited in a generated answer are correlated, not the same outcome, which is consistent with everything above: the AI Overview is drawing on source quality signals that a page's organic position does not fully capture.
What actually gets flagged as a mistake
Google is specific about what triggers a genuine content penalty, and it is method-agnostic. Scaled content abuse is "when many pages are generated for the primary purpose of manipulating search rankings and not helping users," with an explicit example naming generative AI tools used to "generate many pages without adding value for users." Separately, automation itself is not the violation: "If you use automation, including AI-generation, to produce content for the primary purpose of manipulating search rankings, that's a violation of our spam policies." The test in both cases is purpose and value, not whether a person or a tool typed the words. Thin affiliate content (product descriptions copied from the merchant with no original testing or comparison), site reputation abuse (third-party content published on a domain mainly to borrow its existing ranking signal), and expired-domain abuse (buying a domain's history to host low-value content) round out Google's named list, and none of them are about writing craft so much as about what the content is actually for.
Two related practices worth naming directly because they still show up in "content optimisation" advice: cloaking, "presenting different content to users and search engines with the intent to manipulate search rankings," and doorway pages, content built to rank for a cluster of near-identical search queries and funnel visitors to an intermediate page rather than somewhere genuinely useful. Serving one version of a page to a search or AI crawler and a different version to a human visitor is cloaking regardless of the stated justification, and a page built primarily to rank for a keyword variant rather than to answer a real question is a doorway page whether or not it is labelled as content strategy.
Writing to what Google has actually documented, rather than to a word count, a keyword percentage or a schema type that stopped working years ago, is also the writing that the one rigorous study on AI citation found performs best. That is not a coincidence worth ignoring, and it connects directly to the taxonomy question our SEO and GEO guide opens with: content SEO was never a separate discipline from the fundamentals, and neither is writing for citation.
This is the discipline behind the content work inside MSD's SEO and GEO service: checking what a piece of content is actually for before checking how it is structured. If you want a second opinion on whether your own content meets this bar, talk to us.
A few useful questions.
Does Google have an ideal word count or keyword density for SEO content?
Does using AI to write content get a page penalised?
What actually increases the chance of being cited in an AI-generated answer?
Sources and further reading
