--- title: "What AI SEO Actually Means, And Which Parts Are Sold To You Dishonestly" url: https://hostmy.blog/what-ai-seo-means/ date: 2026-09-05 modified: 2026-09-03 lang: en author: "Aditya Sharma" description: "Four claims in this category do not survive contact with the evidence. One of them is a feature we ship ourselves. Here is the honest version." categories: - "RankReady" image: https://hostmy.blog/wp-content/uploads/2026/09/hmb-card-1836-1024x538.jpg word_count: 1597 --- # What AI SEO Actually Means, And Which Parts Are Sold To You Dishonestly AI SEO is a real category with a genuine technical core and a marketing problem large enough to discredit it. The technical core is small, testable and mostly boring. The marketing around it promises outcomes nobody controls, sells files nobody reads, and scores pages against standards that do not exist. Some of that comes from our own category, and one of the weakest claims attaches to a feature we ship. So here is the split, filed under AI SEO because [the argument over GEO, AEO and AI SEO](/geo-aeo-ai-seo/) changes nothing about the work. What AI SEO actually consists of, and which four claims should make you close the tab. AI SEO claims, sorted by whether you can check them AI SEO claims, sorted by whether you can check them Returns a value you can read 200 to GPTBot on a real post URL Markdown served on Accept: text/markdown Vary: Accept present on the response One canonical, one Article node Claim sentence under 18 words Crawler hits counted in the access log Returns nothing you can check Guaranteed citations in ChatGPT An AI visibility score out of 100 Rank one in AI Overviews Plus 300% AI traffic in 30 days The left column is the product. The right column is the pitch. Everything on the left returns a value you can read with curl or find in an access log. Nothing on the right returns anything at all. ## The honest core is four jobs, and all four are checkable Strip the category back and the work is this. **Access.** Can machine clients fetch your pages at all? There is no notice in WordPress when a crawler is refused. A security plugin rule or a host filter can turn agents away for months while your dashboard stays green. **Parseability.** When something fetches the page, can it get the content out without executing JavaScript, and is there structured data describing what the page is? **Shape.** Is there a short, self contained claim near the top of the page, under clean heading order? **Freshness signalling.** Does the page state when it was published and updated, honestly? Every one of those is verifiable with a terminal and ten minutes. None of them requires believing anyone. The hand run version is in [the five layer sweep run by hand with curl](/five-layer-ai-seo-audit/). That is the whole legitimate core. Everything else being sold is either an amplification of one of those four, or one of the following. ## Claim one: llms.txt improves your AI visibility This is the most widely repeated claim in the category and it does not survive the evidence. SE Ranking studied roughly 300,000 domains and found no correlation between having the file and any performance measure. Removing the variable improved their model. Ahrefs checked 137,000 domains, found about 38,000 carrying the file, and 97 percent of those files were never requested once. Google has said it does not use the file. Adoption remains a minority practice. Cyrus Shepard put it plainly: "You don't even need an llms.txt file. At least not yet." Here is the uncomfortable part. Our plugin generates one. So does Rank Math. So does Yoast, as a static file written periodically. A paid WordPress plugin exists whose entire purpose is generating it. The honest reason to generate an llms.txt is that it costs nothing to serve, it is trivially removable, and proposed standards occasionally get adopted late. That is a bet with a small stake, not a feature that does something. Anyone charging you monthly on the strength of it is selling the bet as a result. Our full reasoning is in [why we generate it despite the data](/llms-txt-does-not-work/). ## Claim two: serving Markdown puts your content in front of ChatGPT Content negotiation is real engineering. Your server checks what format the client asked for and answers in that format, stripped of navigation and scripts. The honest limit is narrow. Coding agents send that Accept header. ChatGPT browsing, Claude.ai, Perplexity, Gemini and Grok do not. So the feature is worth having, cheap to run, and it is not a channel to the assistants most people mean when they say AI. Any product describing it as one is describing something that is not happening today. What it actually buys, and the caching trap that most implementations get wrong, is in [serving a plain copy of each post](/markdown-version-of-blog-posts/). ## Claim three: a score out of 100 tells you how AI ready you are Fourteen free readiness checkers exist. Most fetch a URL, look for a handful of files and tags, and return a number. Three problems with that number. There is no published standard for it to score against, so the denominator is invented. Most tools scan one to three URLs, usually including the homepage, and homepages were cited about 4 percent of the time in a large citation sample. And none of the fourteen do passage or sentence level content analysis, so the number describes your container and says nothing about your content. Credit where it belongs. Cloudflare's checker returns a Level from 0 to 5 rather than a percentage, which is a more honest output format, and it runs 20 checks. Search Engine Land's is the only one in the set that queries an actual language model. The full comparison, including where those tools beat ours, is in [every free AI readiness checker compared](/ai-readiness-checker-comparison/). The dishonest version is not the scanning. It is the implied precision of a two digit number. ## Claim four: we will get you cited, mentioned or visible in AI answers This is the one to walk away from. Nobody controls what a model says. Not the plugin authors, not the monitoring vendors, not the agencies. The generation step is not deterministic, the retrieval index is not yours, and the ranking of sources inside an answer is not published. The evidence makes the promise even harder to defend. Only 23.05 percent of cited URLs ranked in the organic top ten, and on ChatGPT that figure was 4.2 percent, so the levers people know how to pull are weakly connected to the outcome, which is how [ranking and being quoted came apart](/ranking-vs-being-the-answer/). Ahrefs, across 863,000 keywords, found AI Overview citations from top ten pages fell from 76 percent to 38 percent. Meanwhile YouTube supplied 9,868 citations and Reddit 6,595, the two largest sources in the sample, and neither is a site you can install a plugin on. The defensible claim is about crawlability and parseability: whether machines can reach and read what you publish. That is a mechanism you can verify. Citations, rankings and visibility are outcomes, and selling an outcome you do not own is where this category loses people. | The claim | What is actually verifiable | | --------- | --------------------------- | | llms.txt improves visibility | No correlation found across 300,000 domains | | Markdown reaches ChatGPT | Coding agents send the header, the assistants do not | | Your AI readiness is 82 out of 100 | No standard exists for that denominator | | We will get you cited | Nobody controls model output | | Your page can be fetched and parsed by agents | Testable in a terminal in ten minutes | ## What is worth doing, ranked by evidence rather than by pitch The work that survives scrutiny is unglamorous and mostly free, and [the WordPress version of that work list](/ai-search-optimization-wordpress/) is four concrete jobs rather than a philosophy. Check access first. A page returning 403 makes every other question theoretical. Then look at your opening, because 41.9 percent of citations came from the first 30 percent of the page. Then fix sentence length, because the mean cited sentence ran 9.27 words across 11,346 cited sentences and nothing longer than 18 words was cited even once, with the 6 to 10 word band carrying 45.2 percent. Then check heading order, since sequential headings appeared 2.8 times more often on cited pages. Two numbers to calm the publishing treadmill. The median cited page was 298 days old, and homepages were cited about 4 percent of the time. This is a job for your archive, not your calendar. The full set is in [the measurement behind the 18 word limit](/18-word-ceiling/). ## What we will and will not say about our own product RankReady is a WordPress AI SEO plugin. It does content negotiation with correct Vary handling, a Markdown version of each URL, per crawler AI controls, AI crawler logging, summary and FAQ blocks near the top of a post, schema output and freshness signals. It runs alongside Rank Math, Yoast, AIOSEO and SEOPress. Tested against all four on WordPress 7.1, there was no duplicate schema, exactly one canonical, and nothing of theirs modified. Setup takes about five minutes. What it changes is whether machines can reach and parse your content. That is the whole claim. It will not get you cited. It will not raise a visibility score that means anything. Its llms.txt generator is a cheap bet, not a feature that works. Its Markdown output is read by coding agents, not by ChatGPT. Everything in the previous four sections applies to us as much as to anyone. You can find it at [RankReady on WordPress.org](https://wordpress.org/plugins/rankready-ai-llm-seo/). ## The test to apply to every vendor in this category Ask what the mechanism is, and ask what would falsify the claim. A mechanism claim sounds like this: your server will return Markdown when the requesting software asks for it, and you can verify that with one command. An outcome claim sounds like this: you will appear in more AI answers. The first can be checked before you pay. The second cannot be checked at all. So the next time a product in this category shows you a number going up, what exactly is it measuring, and who decided the scale?