GEO readiness checker: is your site ready for AI search?

This checklist grades the odds of your page becoming the source of an AI generated answer: whether a crawler can read it, whether it finds a liftable passage and whether it has a reason to prefer you. If what you want to grade is the page's ability to convince a human who just arrived, the right page is the landing page grader, which runs a different model with a different set of 13 items. The two scores measure different things on purpose: you can have a landing page scoring 90 that no AI system ever cites, and a guide scoring 90 on GEO readiness that sells nothing. The first one wins clicks, the second wins mentions.

How to use it

1. Pick one specific page, not the whole site: the guide you want cited, the comparison page, the article that answers your market's question. The score is only actionable at page level.
2. Open yourdomain.com/robots.txt in another tab and look for GPTBot, ClaudeBot, PerplexityBot, Google-Extended and CCBot before ticking the first item. Half of the nasty surprises live in that file.
3. View the page source (Ctrl+U) and search for a chunk of your main text. If it is not there, the content depends on JavaScript and many AI crawlers read an empty shell.
4. Measure the average paragraph length. Paste three paragraphs from the middle of the text into a word counter and take the average: it is quick and it beats an optimistic guess.
5. Tick only what the page actually does today. When in doubt, leave it unticked: a generous diagnosis does not change what the AI reads. Then fill in organic visits and the share of searches in your topic that already show an AI answer, to size the exposure.

How it works: the three layers and the 16 weights

The score is the sum of 16 weighted items adding up to exactly 100 points, arranged in the order an AI system crosses your site. The order matters: a failure in an earlier layer cancels the effort spent on later ones, because a perfect paragraph is worthless on a page the crawler cannot open.

Access is worth 30 because in practice it is binary: either the crawler reads the page or nothing else exists. Citability is worth 40 because it is where almost every site that already does decent SEO loses points, and it is the cheapest layer to fix, since it depends only on how the text was written. Trust is worth 30 because it explains the choice between two equally readable pages, and it is the slowest one to move.

The four numeric items use a ramp rather than a step. Going from 120 to 95 words per paragraph deserves to move the score, otherwise the tool only rewards people who finished the job and gives nobody a reason to start.

Worked example (reproduces the default output)

With the values already filled in, describing a very ordinary corporate blog: robots open, content in the HTML, no gate, sitemap with lastmod, but no llms.txt; question style headings and a table in the body, but no direct answer up front and no schema, with 95 words per paragraph and 1 verifiable fact per section; no original data, no named author, no brand consistency off site, 9 months since the last update and 2 external sources cited.

• Access: 10 + 8 + 5 + 0 + 3 = 26 out of 30, or 87%.
• Paragraph: 8 × (110 − 95) ÷ 50 = 8 × 0.3 = 2.4 out of 8.
• Facts per section: 6 × 1 ÷ 3 = 2.0 out of 6.
• Citability: 0 + 2.4 + 6 + 0 + 2.0 + 4 = 14.4 out of 40, or 36%.
• Freshness: 7 × (18 − 9) ÷ 15 = 7 × 0.6 = 4.2 out of 7. Sources: 5 × 2 ÷ 4 = 2.5 out of 5.
• Trust: 0 + 4.2 + 0 + 2.5 + 0 = 6.7 out of 30, or 22%. That is the weakest layer.
• Score: 26 + 14.4 + 6.7 = 47.1, rounded to 47 out of 100, band D, the AI cites someone else.
• With 12,000 organic visits per month and 35% of the topic's searches already showing an AI answer: 4,200 exposed visits. 4,200 × 53 ÷ 100 = 2,226 visits per month at risk.

The fix queue comes out like this: move the answer to the top (+10), publish original data (+7), add the schema (+6), sign it with an author (+6), shorten the paragraphs (+5.6), standardise the brand off site (+5), publish llms.txt (+4), add verifiable facts (+4), refresh the content (+2.8) and cite more sources (+2.5). Those ten add up to 52.9 points, which is the gap of 53 before rounding. That is what the tool shows when the page loads, and it is the same reasoning you repeat with your own numbers.

How to read it, and where the score misleads

The score measures readiness, not results. It tells you whether you cleared the known reasons for being dropped, and nothing beyond that. Domain authority, what a competitor published the same week and the index each system used that day stay outside your control and keep deciding a large share of citations.

The second trap is self deception while answering. Almost everyone ticks "direct answer up front" by reflex, because they wrote the text and know where the answer is. The honest test is to read only the first paragraph out loud and ask whether it answers the question on its own, without the rest of the page. If it needs context, untick it. The cost of lying here is optimising the wrong layer for a whole quarter.

The third is treating visits at risk as a forecast of loss. The number shows how much of your current exposure is subject to becoming an answer with no click, with the page as it stands, scaled by the size of your gap. It is an order of magnitude for sizing effort, not a prediction: some of those searches would never have produced a click anyway, and some people will keep clicking even with the answer on screen. The long version of that argument is in zero-click search optimization.

Finally, the model treats the 16 items as independent, and they are not. Rewriting the opening as an answer usually shortens the paragraph and makes the schema easier too, so the gains do not add up cleanly. One more reason to fix by layer and watch what happens to traffic after each wave.

From checklist to test

A checklist is not evidence. Every item in your queue is a hypothesis about how a third party system treats your page, and the only honest way to know is to measure what changes. The short path is this:

• Start with the weakest layer the tool flagged, not with the item you find most fun. A layer at 22% of its points is where the money is sitting still.
• Record today's score and the date. Without a baseline, any later traffic swing turns into free interpretation.
• Apply one wave of fixes at a time, across a group of similar pages, and leave a matching group untouched. It is as close to a control as you get when the outcome depends on a third party.
• Track mentions and citations, not just rankings. The method is in the generative engine optimization guide, and the FAQ structure that earns the most citations is in FAQ content for AI citation.
• When the change also touches conversion, measure it with an A/B test before rolling it across the site: a short answer at the top helps the AI and sometimes hurts the sale. The relationship between the two is covered in does A/B testing affect AI citations.

To turn the first item of the queue into a testable hypothesis, use the hypothesis generator in the if, then, because, measured by format. To size the experiment before switching anything on, the sample size calculator settles it in a minute.

Frequently asked questions

What is GEO, or generative engine optimization?

GEO is the work of making your content findable, understandable and citable by systems that answer instead of linking: the AI summary at the top of search, a chat assistant, an answer engine. The practical difference from classic SEO is the target: SEO competes for a slot in a list of ten links, GEO competes to be the source a system lifts inside a single answer. Much of the work overlaps, because nearly every answer system starts from a search index, but what wins the citation is different: an extractable passage, a verifiable fact and an authorship signal matter more than an optimised title.

Does this tool crawl my URL automatically?

No, and that is a design choice. Reading your page would require a server of ours fetching your site, and even then it could not answer the questions that weigh most here: whether the direct answer comes early, whether the data is yours or copied, whether the author has a credential. Those three items are worth 23 of the 100 points and only someone who knows the content can answer them honestly. You spend three minutes and get a better diagnosis than any crawler would produce. Nothing you type leaves the browser: the page makes no network calls at all.

Should I allow or block AI crawlers in robots.txt?

It depends on your business model, and this checklist assumes you want to be cited. If your content is the bait that creates demand, blocking GPTBot or ClaudeBot removes you from the answer and lets a competitor answer in your place. If your content is the product people subscribe to, blocking is legitimate defence and you should treat that item as intentional. The common mistake is not the choice, it is not noticing: plenty of sites sit blocked through an inherited template or a security plugin, with nobody having decided anything.

What counts as a good GEO readiness score?

From 85 to 100 the page is ready to be cited, from 70 to 84 it is close, from 55 to 69 it is readable but forgettable, from 40 to 54 the AI has reasons to cite someone else, and below 40 it is effectively invisible to AI search. Most sites that already do decent SEO land between 45 and 65, almost always through the same pattern: access is fine and structure is reasonable, but the text was written to be read end to end, with no direct answer up front and no original data anywhere.

Why is the citability layer worth more points?

Because it is where almost everyone loses, and it is the only one of the three that depends purely on how you write. Access is usually fine by accident, since the site has to be crawlable for Google anyway. Trust is built over months. Citability can be fixed this week: move the answer into the first paragraph, break the text into self-contained blocks, turn the comparison into a table. It is the layer with the best return per hour of work, which is why it carries 40 of the 100 points.

Does a high score guarantee the AI will cite me?

No. No system publishes its real source selection criteria, and citation depends on things outside your page: domain authority, what competitors published, how the question was phrased, which index the system used that day. What an honest checklist does is clear the known reasons for being dropped. That is the difference between not being considered and being considered: a high score buys entry to the race, not the result of it.

Keep going

With the score and the queue in hand, the next step is to pick one layer and work it to the end. Read the full method in the GEO guide, see what changes in conversion work in CRO in the age of AI, and grade the same page for persuasion with the landing page grader.