Can AI read your website? Here is how to fix it.
Your next customer is asking ChatGPT who to hire. Before it answers, it reads your website. The AEO Fix Guide shows you exactly what it is looking for, in the order that moves your score the most.
- A robots.txt that lets the AI search bots in
- A fill-in-the-blanks llms.txt, the “one text file” that took jmanai.com to a 96
- A ready-to-paste JSON-LD block for your type of business
- Homepage formulas, a before and after, and the prompt that makes your AI do the typing
Get the AEO Fix Guide
What is inside
Built on the same 28 checks the free AEO Site Checker scores, so every page maps to points you can see on your own report.
What every check is actually testing
The guide above is the short version you can print. Below is the long version, free and in full: all 28 checks, what each one is looking for, and which ones are worth paying somebody to do rather than fighting with yourself on a Sunday.
How the score works
Every check is worth a fixed number of points, 110 in total. Contact signals only apply to pages about a business or a person, so a blog post is not penalized for having no phone number. Your score is the points you earned divided by the points that applied to your page, times 100.
One thing worth knowing before you read further: the checker scores a single page, not your whole site. Running your homepage tells you about your homepage. If you want to know why nobody finds your Brighton listings, check the Brighton page.
Why the categories are weighted so unevenly
The six categories are not worth the same, and the gap is deliberate. Fetchability alone is 40 of the 110 points, roughly 36 percent of everything available, because it is binary. If an AI crawler cannot load your page, nothing else on this list can save you. The checker puts it bluntly when you fail it: “A blocked page is cited exactly as often as a blank one.”
Content quality, the thing most people want to talk about, is worth 6, which is about 5 percent of the total. Not because writing does not matter, but because these four checks are proxies for good writing, not measurements of it. A machine can tell whether your opening paragraph answers a question. It cannot tell whether the answer is any good. So the checker weights what it can actually see and leaves the rest to you.
Work down this page in order. The categories are already sorted by what will move your number most.
1. Fetchability
40 points, 5 checks
Can a machine open the door at all. This is the only category where a single failure can sink everything else.
Direct fetch as an AI search bot
16 pts
Depends where your site lives
The checker requests your page while identifying itself the way OpenAI’s search crawler does. If your server hands the page over, you get all 16 points. If a browser can load the page but the bot gets turned away, something between you and the internet is refusing robots on sight. Usually that is Cloudflare, a WAF, or a bot-protection setting your host turned on to stop scrapers.
This is the single largest number on the scorecard and the most common silent failure, because the site looks perfect to you. You are using a browser. The crawler is not.
If you control Cloudflare or your host’s security settings, this is a checkbox and takes five minutes. If you are on a vendor platform that runs your site for you, which covers most brokerage-provided agent sites, you cannot fix it yourself and have to ask them to allow verified AI search crawlers. Ask in writing, name the crawlers, and expect to explain why.
robots.txt allows AI search crawlers
10 pts
Do it yourself, 10 minutes
robots.txt is a plain text file at the root of your site that tells crawlers what they may read. The checker looks at nine crawlers that actually produce citations, including OAI-SearchBot, Claude-SearchBot and PerplexityBot.
Here is the distinction almost everybody gets wrong. Search crawlers are the ones that fetch your page when a person asks a question, and they are how you get named in an answer. Training crawlers, like GPTBot, ClaudeBot and CCBot, collect text to train future models. Those are a separate decision, and blocking them costs you no citations at all. If you have strong feelings about your writing being used as training data, block the training crawlers and leave the search crawlers alone. A lot of sites blocked all of them in one angry afternoon in 2023 and have been invisible ever since.
No robots.txt at all is fine. It means nothing is blocked.
Content is in the HTML, not a JavaScript shell
8 pts
Depends on how the site was built
Your browser runs JavaScript, so it can assemble a page out of nothing and show you a finished website. Most AI crawlers do not run JavaScript. They read the raw file the server sends and nothing more. If that file is basically empty and the words arrive later, the crawler sees a blank page.
Most website builders handle this correctly on their own. The sites that fail are custom builds, usually React or a single-page app, where nobody asked for server rendering because nobody was thinking about crawlers.
There is no setting for this on a custom site. It is a development job, and depending on how the thing was built it can be either an afternoon or a rebuild. Get a straight answer on which before you agree to anything.
Served over HTTPS
4 pts
Do it yourself, usually free
The padlock. Every serious host includes a certificate at no cost and turns it on by default. If you are failing this one in 2026, something is genuinely misconfigured, and it is worth fixing in the next hour for reasons that have nothing to do with AI.
HTML size under 1.5 MB
2 pts
Do it yourself, low priority
Enormous pages get truncated or skipped. This is almost always caused by page builders inlining every script, style and icon into the document itself. Worth two points, so fix it when you are already in there, not as its own project.
2. Answer Engine signals
25 points, 7 checks
The things engines read that Google never asked for. This is the category where most people have the most points sitting on the table.
JSON-LD structured data
6 pts
Worth paying for
A block of machine-readable code that states plainly who or what this page is about: name, phone, address, area served, website, and links to your profiles elsewhere. Humans never see it. It is the difference between a machine guessing you are a real estate agent in Rochester and a machine knowing it.
This is the highest-value item in the category and the one I would least recommend hand-rolling. Malformed schema is worse than none, because it can describe you incorrectly and the engines will believe it. If you have a developer, this is an hour of their time. If you do not, a decent SEO plugin will generate a basic version, which is much better than nothing.
Extractable article body
5 pts
Do it yourself, an afternoon of writing
At least 300 words of real paragraphs that a parser can lift. Not headlines, not captions, not a hero image with six words on it. Engines cite text. A beautiful page with 40 words on it has nothing to cite.
Write who you serve, what market, and what actually makes you different, in the voice you would use out loud. This is free, it only costs you an afternoon, and it feeds three other checks further down this page at the same time.
llms.txt present
4 pts
Do it yourself, 20 minutes
A short markdown file at the root of your site that names what the site is, what it covers, and links to the pages that matter most. Think of it as a hand-written index for machines.
It is a newer convention and not every engine uses it yet. It is also four points for twenty minutes of typing and a file upload, which makes it the best effort-to-points trade on this entire page. Cheap insurance.
Site breadth
4 pts
A strategy, not a fix
How many distinct internal pages this page links to. A single page rarely gets cited, because a site with one page does not look like an authority on anything.
You cannot fix this in an afternoon and you should not try. It is the slow one: neighborhood pages, service pages, a blog that actually gets written. If you are sitting at one or two pages, this check is telling you something true about your whole site rather than about this page.
Author and organization provenance
3 pts
Bundle it with the schema
Two things together: a named author or publisher in your structured data, and sameAs links pointing to your profiles on LinkedIn, Instagram, YouTube, your brokerage page and anywhere else you exist. It is how an engine confirms that the person on this page is the same person it has seen elsewhere.
Whoever writes your JSON-LD should do this in the same sitting. Doing it separately is doing the same job twice.
Content freshness
2 pts
Do it yourself
A dateModified in your structured data, updated when the page actually changes. Engines weight recency, and a page with no date at all is treated as undated rather than new.
One warning. Do not automate this to touch every page nightly so everything looks fresh. It is transparent, and a page that claims to change daily while saying the same thing for two years is a worse signal than an honest old date.
llms-full.txt present
1 pt
Optional
A single file containing the full text of your key pages. One point. Genuinely optional. Do it if you enjoyed making llms.txt, skip it otherwise.
3. Core SEO
21 points, 7 checks
The old fundamentals. Google AI Mode is still search underneath, and most of these are ten-minute jobs you have been meaning to do anyway.
Title tag
4 pts
Do it yourself, 10 minutes
Between 15 and 70 characters. Your name, your market, what you do. Too short and it says nothing, too long and it gets cut off mid-thought.
Meta description
4 pts
Do it yourself, 10 minutes
Between 80 and 175 characters stating the market and the service in one sentence. This is also the sentence that shows up under your link when somebody shares it, so it earns its keep in more than one place.
XML sitemap
4 pts
Usually a plugin
A machine-readable list of every page on your site, published at a predictable address and referenced in robots.txt. On WordPress this is a checkbox in any SEO plugin. On most builders it is automatic. Four points for something you may already have and simply never pointed at.
Canonical URL
3 pts
Check it, do not assume it
A tag saying which address is the real one for this page, so the same content at several addresses does not compete with itself.
Most platforms set this automatically, which is exactly why it is worth checking rather than assuming. Vendor-built agent sites are notorious for canonicalizing every page to a corporate address that does not exist, which hands your credit to somebody else’s 404.
Open Graph tags
3 pts
Do it yourself, plugin or 15 minutes
og:title, og:description and og:image. These control the preview card when your link gets pasted into Facebook, LinkedIn or a text message. Without them the preview is whatever the platform scrapes, which is usually your logo and a shrug.
Twitter card
2 pts
Do it yourself, same sitting
The same idea for X. Two points, and if you are already adding Open Graph tags it is the same three minutes.
Language declared
1 pt
One attribute
A lang attribute on the html element. One point, one word, and it matters more than one point suggests if you publish in more than one language.
4. Semantic HTML
13 points, 4 checks
The skeleton a parser uses to tell your actual answer apart from your navigation, your cookie banner and your footer.
Semantic landmarks
4 pts
Usually a theme-level job
Real header, nav, main, article and footer elements rather than a stack of anonymous divs. Without them a machine has to guess which part of the page is the content, and it guesses wrong often enough to matter.
This lives in your theme or template, not in your page content, which is why you usually cannot fix it from the page editor. It is a real but small development task, and it fixes itself across every page at once, which makes it better value than it looks.
Exactly one H1
3 pts
Do it yourself, 10 minutes
One top-level heading that states the page’s answer: who, where, what. Page builders love to hand out H1s like candy, so it is common to find four of them, or none at all because the headline was styled as a paragraph.
Heading hierarchy
3 pts
Do it yourself
Do not jump from an H1 to an H3. Headings are an outline, and skipped levels break the outline. This almost always happens because somebody picked a heading level based on how big the text looked rather than what it meant.
Image alt text
3 pts
Do it yourself, tedious but easy
Describe every meaningful image in plain words: what it shows and where it is. Not a keyword list. A real sentence a person would say.
Decorative images do not need it. Your listing photos absolutely do, and this is the one check on the page that also directly helps somebody using a screen reader, which is reason enough on its own.
5. Content quality
6 points, 4 checks
Only six points, and the two that replicated in the Princeton research on generative engine optimization are the first two. Small numbers, real effects.
Front-loaded answer
2 pts
Do it yourself, and do it first
Start the page with two or three sentences that answer the question somebody actually came to ask. Who you are, what you do, for whom, where. Engines lift the opening far more often than the middle.
Two points, but this is the highest-leverage writing change on the entire page, because it is also the paragraph a human decides on. Most sites open with a slogan. A slogan answers nothing.
Statistics and numbers
2 pts
Do it yourself, 15 minutes
Concrete figures: years in business, homes sold, average days on market, review count. Numbers get cited because they are checkable and specific, and a sentence with a real number in it survives being lifted into an answer better than one without.
Question-style headings
1 pt
Do it yourself
Headings phrased the way people ask: how, what, why, when, where, who. It matches the shape of the query, which makes your section easier to match to the question.
Quotations
1 pt
Do it yourself
Real quotation marks around something somebody actually said. A client quote or two. Quotations lift citations, and you almost certainly have a folder of them you have never put on the page.
6. Contact signals
5 points, scored only when it applies
Skipped entirely for pages that are not about a business or a person, so your blog posts are not punished for having no phone number.
Contact signals
5 pts
Do it yourself, 20 minutes
Phone, address, hours and service area, in the footer as text and in your structured data as telephone, address, openingHours and areaServed. Both places. The footer is for people, the structured data is for machines, and skipping either one costs you.
Phone numbers baked into an image do not count, for you or for anybody trying to tap it on a phone.
So what do you actually do first?
If you only do four things, do these, in this order.
Confirm an AI crawler can load the page, because 16 points and every other point depend on it. Check robots.txt for a blanket block you may have inherited, which is another 10. Publish an llms.txt, which is four points for twenty minutes. Then rewrite your opening paragraph so it answers the question instead of announcing a slogan.
That is roughly 30 points and one afternoon, and three of the four are free.
What is genuinely worth paying for
Three things on this page resist do-it-yourself, and they are the structured data, the semantic landmarks, and server rendering if your site turns out to be a JavaScript shell. The first two are small, well-defined jobs that a competent developer finishes in a morning. The third is the one to get a real estimate on before you agree to anything, because the honest answer is sometimes an afternoon and sometimes a rebuild.
Everything else on this page is yours to do. Most of it is typing.
Then re-run it
Fix the top three, run the check again, and watch the number move. That loop is the entire point of the tool, and it is the step almost everybody skips. Nothing here is permanent either. Sites get redesigned, hosts turn on bot protection without telling anybody, and a page that scored an A in March can quietly fail in June.
Rather learn this step by step?
The on-demand course AI Is the New Front Door walks you through all of it on video: scoring your site, the 3 A’s of AI visibility, fixing your bio and Google Business Profile, and rebuilding a listing so AI can match it to a real buyer. Twelve short lessons, a handout, and a certificate when you finish. $47.
Want AI to sound like you while you’re at it? Pair it with Me, Myself & AI, also $47.