How to use this list
Work top to bottom and stop at the first thing that fails, because the checks below it cannot help you until it passes. This is a sequence rather than a scorecard, and treating it as a scorecard is how businesses end up commissioning content while a firewall rule refuses every request.
The ordering is by two things at once: how binding the check is, and how little it costs to fix. Those correlate more than you would expect. The failures that cost you everything are usually one-line settings, and the failures that cost you a little are usually quarters of work.
Nothing here requires a tool. All fourteen can be run by a person with access to the site and an afternoon, which is deliberate. The point of a checklist is to establish where you stand before anybody is paid to tell you.
Write down what you expect to find. Teams that predict a clean pass and then fail check two tend to take the result more seriously than teams that went in expecting problems.
Checks 1 to 3: can anything reach you
Most sites that are entirely absent from AI answers fail one of these three, and all three are settled in an afternoon.
1. Are answer crawlers permitted in the rules on your site. Not "is the site crawlable". Find the file, read it, and list which named clients are allowed and refused. Search crawlers and answer crawlers are different clients, and a site can be wide open to one while closed to the other.
2. Are they permitted at the edge. This is the separate one people miss. A permissive file on the site with a restrictive rule in front of it produces a refusal the site never sees. Bot management, firewall rules and hosting defaults all live here, and they are usually set by somebody who was solving a different problem.
3. Does the page return its content without running scripts. Fetch the page as a plain document and read what comes back. If the words you care about are not in that text, they may not be available to anything that is not a full browser. This is the check that catches an entire category of modern site build.
If any of the three fails, fix it and wait for pages to be refetched before doing anything else. A great deal of expensive work has been commissioned by companies whose only real problem was on this list.
Checks 4 to 7: can a passage be used
These are editorial and they are the highest-return work on most sites, because commercial writing has habits that happen to be hostile to extraction.
4. Do your ten most important pages answer their own title in the first two sentences. Read only the opening. If the answer is further down, move it up. You are not writing anything new, you are relocating a conclusion you already reached.
5. Do your claims survive being removed from the page. Look for sentences that start with "this", "it" or "they", comparisons with an unstated other half, and figures without their unit. All of them read perfectly in place and become meaningless when lifted.
6. Is the answer in prose rather than only in interface. Specifications inside tabs, terms inside accordions, sizing inside a modal. Keep the summary in plain text and let the interface hold the detail.
7. Do headings describe content rather than tease it. A heading that says what the section establishes helps a reader skim and helps anything else find the relevant passage. A heading written as a hook helps neither. Our page on writing for LLMs goes into what a liftable paragraph looks like in practice.
Checks 8 to 11: is there enough of you
Now the slower half. These cannot be fixed this week and they are the reason a technically perfect site can still be invisible.
8. How many pages genuinely cover your priority subject. Not pages that mention it. Pages that answer a distinct question within it. Count them honestly. One is a page, and a page is thin evidence that you are the thing to reach for.
9. Are those pages connected to each other. A folder of pages that do not link to one another is a filing decision rather than a body of work. Each should point at the two or three a reader would genuinely want next.
10. Is anything on the site out of date in a way that undermines it. A confident claim attached to an old date is worse than no claim. This matters more than usual here because a passage gets lifted without its context, including without the reader noticing when it was written.
11. Is your category discussed anywhere you do not own. Search for the way a buyer describes the problem and see whether the discussion exists and whether you are in it. There is no lever for this and it is still worth knowing, because it tells you whether your absence is a site problem or a presence problem.
Checks 12 to 14: can you tell if it worked
The last three are about measurement, and skipping them means everything above becomes unfalsifiable.
12. Do you have a fixed set of questions. Between six and ten, written the way a buyer would phrase them rather than the way you describe yourself. Fixed is the operative word. A set that changes between readings cannot show movement.
13. Do you ask each one more than once. The same question put twice can return different names. A single pass tells you what happened once, which is not the same as what is normal. Several readings tell you a distribution, and a distribution is the thing that can move.
14. Do you have a reading from before the work started. This is the one that cannot be recovered later. If you make changes and then start measuring, you will spend the next quarter unable to separate what you did from what the assistant did on its own. Take the reading first, even a rough one. Our walkthrough on tracking mentions by hand is enough to do it without buying anything.
What actually fails most often
In practice the failures cluster, and they do not cluster where teams expect.
The commonest total failure is check two, the edge rule nobody wrote deliberately. It produces complete absence with no symptom, and because it is invisible from inside the site it survives audits that only read the site's own files.
The commonest partial failure is check four, the opening that circles. It rarely produces total absence. It produces a site that gets fetched, considered and passed over in favor of something that stated its point sooner, which is a much harder failure to notice because nothing is broken.
The commonest skipped check is fourteen, the baseline. It is skipped because it produces nothing to show anybody and takes two weeks to be worth having. It is also the one that makes every other item on this list judgeable, so skipping it converts a program of work into a matter of opinion.
The check teams most want to jump to is eight, depth, because it is the one with a budget attached and a plan that looks like progress. It is genuinely important and it is fourth in line for good reasons.
The honest trade-off
A checklist is a blunt instrument and it is worth saying where it misleads. It implies every item is independent and equally weighted, and neither is true. Checks one to three are gates, and checks eight to eleven are a slope. Passing all fourteen is not a state anybody reaches, because eleven has no finish line.
It also cannot tell you whether the subject you chose is worth owning. A site can pass every check on this page and remain invisible because it has committed to a subject nobody asks about. That is a demand question and no visibility checklist answers it.
What the list does well is stop you spending a quarter on the slow half while the fast half is broken, which is the single most expensive mistake available in this work.
One further limit is worth stating. Nothing here measures whether being named actually produced anything. A person who reads your name in a generated answer and then searches for you directly arrives looking like ordinary direct traffic, so the assistant that caused the visit leaves no fingerprint. Every item on this list is about being present rather than about what presence is worth, and conflating the two will eventually produce a disappointing conversation with somebody holding a budget.
Three ways this list gets misread
The checklist format invites three specific mistakes, and all three are worth naming because they are how the exercise gets run and still achieves nothing.
Treating it as a score. Eleven of fourteen is not a grade. If the three you failed are checks one to three, you are entirely absent from AI answers and the eleven passes are irrelevant. The first three are gates and the rest are improvements, and averaging a gate with an improvement produces a number that means nothing at all.
Running it once. Access rules change without announcement, because they are changed by people solving unrelated problems who have no reason to think about this. A site that passed in spring can be refusing every request by autumn with nobody having decided anything. The first three checks want to be automated so a change surfaces on its own rather than waiting for somebody to remember.
Delegating it to whoever is free. Checks one to three need somebody with real access to the site and the edge, and they need to read the rules rather than ask about them. An answer collected by asking a colleague what the setting is will confirm the intention rather than the configuration, and the whole point of the check is that those two have come apart.
The version of this that works is short, evidenced, dated, repeated, and run by somebody who can see the actual files. Everything else is a workshop.
How to record what you find
The checklist is only worth running if the result survives the week, and in most companies it does not, because it is run in somebody's head during an afternoon and reported as a summary.
Write down, for each check, the answer and the evidence rather than a verdict. For access that means the actual rule text and where it lives, not "crawlers allowed". For the opening-sentence checks it means the page and the current first two sentences, not "needs work". Evidence survives a handover and a verdict does not, and this work reliably outlives the person who first ran it.
Date every entry. A check that passed in March and a check that passed last week are different facts, and access rules in particular change without anybody announcing it, because they are usually changed by somebody solving an unrelated problem.
Keep the failures visible after they are fixed rather than deleting the row. Knowing that the edge rule was wrong once tells the next person where to look first when the symptom reappears, and it will reappear, because the conditions that produced it the first time are usually still in place.
The version of this that actually holds is a short table with a row per check, a column for the answer, a column for the evidence and a column for the date. Anything more elaborate stops being maintained by the second quarter, and anything less is a memory.
Bottom line
Run the first three checks today. They take an afternoon, they need no tooling, and they resolve the most common cause of complete absence. If any of them fails, stop and fix it before reading further, because nothing below can help you until it passes.
Then take the baseline, even a rough one, before you change anything. Everything else on the list becomes an argument rather than a measurement without it, and it is the only item here that cannot be produced after the fact.
Do the editorial checks next because they are fast and they improve the pages for people too. Leave depth until last, not because it matters least but because it is the only one that will still be true in three months if you get the order wrong.