--- title: "E-E-A-T for AI Answers Is Judged One Sentence at a Time" url: https://hostmy.blog/eeat-seo-ai-answers/ date: 2026-09-17 modified: 2026-09-03 lang: en author: "Aditya Sharma" description: "Google evaluates E-E-A-T across a whole site. An AI answer evaluates the sentence it is about to quote. That changes the work." categories: - "AI Search" image: https://hostmy.blog/wp-content/uploads/2026/09/hmb-card-1872-1024x538.jpg word_count: 1677 --- # E-E-A-T for AI Answers Is Judged One Sentence at a Time A consultant I work with has twenty years in her field, a real byline, an about page with her credentials, and a site that Google clearly trusts. She ranks. She sends me a message in February asking why AI assistants describe her competitor as the expert and mention her not at all. Her E-E-A-T, in the Google sense, was excellent. It was doing nothing for her in AI answers, and the reason took me a while to see. Google evaluates trust across a domain and then decides where to rank a page. An AI answer does something narrower. It picks a sentence, puts it beside sentences from other sources, and needs that specific line to hold up on its own. Authority stops being a property of your site and becomes a property of each claim you make. ## The unit of judgement got much smaller In an analysis of 11,346 cited sentences a May 2026 study could extract, the mean cited sentence was 9.27 words. The median was 10. Nothing above 18 words was cited a single time. So whatever authority you have has to survive compression into about ten words, stripped of the paragraph that qualified it. [The measurement behind that 18 word ceiling](/18-word-ceiling/) is worth reading once, because it changes what a good sentence looks like. Try this on your own writing. Take the most authoritative claim on your best page and cut it down to ten words with no context around it. If what remains is still checkable and still specific, it can travel. If what remains is "our approach delivers better outcomes for clients", it cannot, and no amount of site-level trust rescues it. That is the difference in one exercise. ## What each signal actually does, on each surface Most E-E-A-T advice is written for Google's raters. Some of it transfers and some of it is doing nothing for you in an AI answer. | Signal | Effect on Google | Effect in an AI answer | | ------ | ---------------- | ---------------------- | | Detailed about page | Site-level trust | Weak, homepages and about pages are cited around 4 percent of the time | | Author bio with credentials | Meaningful | Meaningful, but only if the credential sits near the claim | | Named source next to a number | Helpful | Strong, gives the model something corroborable | | Vague expertise language | Neutral | Useless, nothing to quote | | Outbound citations | Helpful | Helpful, and checkable against other sources | | Long nuanced paragraphs | Good for readers | Above the 18 word ceiling, not quotable | | Case study with real figures | Good | Strong, specific and unusual | The row that surprises people is the about page. Homepages are cited around 4 percent of the time, which is [why the homepage is the wrong page to optimise](/homepage-wrong-page-for-ai/), and the effort many sites pour into an authority-signalling front door is effort not spent on the article pages carrying the rest of the opportunity. ## Experience shows up as detail, not as adjectives The first E in E-E-A-T is experience, and it is the one that survives compression best, because real experience produces specifics and specifics are quotable. Compare two ways of saying the same thing. > Our team has extensive experience making WordPress sites easy for AI crawlers to read. > A 4.7 MB page came down to 18 KB of Markdown, identical words, same page. The first is 14 words and contains nothing a model can use, because it asserts a quality rather than a fact. The second is 15 words and contains two numbers somebody can check against their own pages. Both are inside the length ceiling. Only one is worth quoting, and [writing claims short enough to be lifted](/write-sentences-ai-can-quote/) is a craft skill more than a technical one. This is why the standard advice to "show your experience" fails in practice. People add adjectives. Experience is evidenced by numbers, dates, edge cases and things that went wrong, and those are exactly the details generic content cannot fake. Two ways to write the same claim Two ways to write the same claim Survives being quoted Backups run hourly and are kept 30 days. A 4.7 MB page came down to 18 KB of Markdown. GPTBot returned 200 on every post URL. The median cited page was 298 days old. Does not We are committed to quality. - Industry leading performance. Our team has years of experience. Best in class support. Mean cited sentence, 9.27 words. Nothing above 18 words was cited even once. Experience is judged one sentence at a time. A sentence with a number in it survives being lifted out of the page. A sentence about commitment does not. ## Authority is corroboration, and it mostly lives off your site An awkward finding, but it explains my consultant's problem better than anything on her own site did. The two largest citation sources measured are YouTube, with 9,868 citations, and Reddit, with 6,595. Neither is a company blog. When a model builds a picture of who is authoritative on a topic, a large share of the material comes from places you do not own, which is [why forum threads and video transcripts get cited ahead of blogs](/why-ai-cites-reddit-youtube/). Her competitor was not a better writer. He answered questions in a niche forum for four years and had given eleven conference talks that were on YouTube with transcripts. That trail exists independently of his website, and it agrees with his website. What worked for her, over about five months, was unglamorous: Answering real questions properly in the two communities where her clients already are, with specifics and without a link unless asked. - Getting her talks transcribed and published rather than left as slide decks. - Making sure the claims on her site matched the claims she made everywhere else, because disagreement between sources is visible in a way it never was on a results page. What can be verified is that her claims are now specific, consistent across sources, and stated in sentences a machine can lift. Whether that produced citations is not measurable from the outside, and anyone who tells you it is has not tried. ## Trustworthiness now means being checkable The T changed the most, and it changed in your favour if you write honestly. When an assistant assembles an answer, it places short statements from several sources next to each other. Contradictions become obvious. A number that appears on four sites and disagrees with yours makes your line the odd one out. That gives you three practical rules. **Date your claims.** "As of August 2026" is four words and makes a stale figure defensible rather than wrong. **Name your source in the sentence.** "Ahrefs studied 863,000 keywords" carries its own provenance. In their study of 863,000 keywords, AI Overview citations from top-ten pages fell from 76 percent to 38 percent, which is a claim you can check without trusting me. **Correct in public.** An updated page with a visible correction is stronger than a page that quietly changed. It also keeps the page from going stale, and stale pages are roughly three times more likely to lose citations they previously held. ## Freshness is a trust signal, and most sites fake it The median cited page is 298 days old, which surprises people who assume recency wins. Ten month old pages are the norm among cited content. But stale pages lose ground at about three times the rate. Old is fine. Abandoned is not. The failure I see constantly on client sites is the fake update. A plugin bumps the modified date, the page displays "Updated August 2026", and not a word changed. That is a trust claim you cannot support, it is visible to any reader who has seen the page before, and it does nothing for the underlying problem. A real update means checking the facts, fixing what is wrong, and saying what changed. Six of those a year beats sixty new posts, and it is less work. ## What this means for a WordPress site specifically Three of these are structural rather than editorial, and WordPress does not do them for you by default. Author identity has to be attached to the content, so a model reading the page can see who made the claim. A real author box near the content beats an about page two clicks away. Dates have to be honest and machine readable. Published and modified dates need to be in the markup, not only in your theme's display, which is one job among the [Article, FAQPage and Person markup a WordPress site needs](/structured-data-wordpress/). Claims have to be in a structure that can be parsed cleanly. Sequential headings, H2 then H3 then H4 without skipping, correlate with a 2.8x lift, and fewer sections placed well beats many. That structural half is what [RankReady](https://wordpress.org/plugins/rankready-ai-llm-seo/) is for. It is a WordPress AI SEO plugin that handles the author box, honest freshness signals, schema and FAQ output, summary blocks near the top of a post, and a Markdown version of each page for machine readers. It runs alongside Rank Math, Yoast, AIOSEO and SEOPress rather than replacing them, so your titles and sitemaps stay exactly where they are. Setup takes about five minutes. What it makes possible is a page that is fetchable and parseable with the authorship attached. The credibility of the sentences inside it is still your job, and no plugin will ever change that. ## Start with one page and one claim Pick your single most important page. Find the one claim you most want repeated about you. Then ask three questions. - Can it be said in under 18 words without becoming vague? - Does it contain a number, a date or a name that someone could check? - Does it appear in the first 30 percent of the page, where 41.9 percent of citations come from? Three noes means the claim is not travelling anywhere, regardless of how much authority your domain has accumulated. The habits that carry over from ordinary SEO, and the ones that quietly stopped working, are sorted in [what actually changes with ChatGPT SEO](/chatgpt-seo-what-changes/). To see what is currently being said about you, [the quick test of what it knows about you](/fifteen-second-brand-test/) is the fastest read available. What is the one sentence you would want an assistant to repeat about your business, and is that sentence written down anywhere on your site?