# machine layer > machinelayer.org — being found, understood and believed by AI systems A record of making one website work for readers that are not people. Being reached, being parsed and being believed are three separate problems with three separate fixes, and one of them turned out to be solved already. ## what this is A method and a single worked example. 31 entries, each written to stand alone, since any one of them may be the only page someone sees. Claims are not marked; the observations behind them are. Each observation carries how far it was established — replicated, seen more than once; single, seen once; mechanical, a property of how retrieval works rather than a trial result; unclassified, where replication was never established. There are 28 in all, 9 of them replicated. ## what this is not Not a service. Nothing is sold here and there is nothing to sign up for. Not a general rule of the web either: the evidence is one site, across four providers and both consumer and developer surfaces, over seventeen versions of its machine-facing copy — and it is described as such throughout. ## the subject Every observation was made on fastfoodvisors.xyz, a small unmonetised exhibit with no purchase, no account and no sign-up. That absence is what makes it a clean subject — none of the results are confounded by a funnel. The site began with an unusual top-level domain, a pseudonymous author, vocabulary that trips automated risk heuristics, and no reputation of any kind, which is the point: the method is demonstrated under the worst starting conditions rather than the easiest. The subject is not independent. This record was written by the person who runs that site, so nothing here is a third party's assessment of anything. It is the reason every observation carries its status and the subject itself stays public: a measurement can be re-run by someone else, and a conclusion has to be taken on trust from whoever wrote it down. ## pages ### reach — Did the system get your page at all? - [A cautious answer usually means nothing was found](https://machinelayer.org/a-cautious-answer-usually-means-nothing-was-found): The most misread signal in this whole area. A hedging, non-committal answer feels like a verdict on your site. It is almost always a verdict on an empty… - [Similar names can inherit each other's reputation](https://machinelayer.org/similar-names-can-inherit-each-others-reputation): If your domain resembles other domains, a system that fails to find you may answer from them — and hand you their reputation. - [The chatbot and the API do not fetch the same way](https://machinelayer.org/the-chatbot-and-the-api-do-not-fetch-the-same-way): Retrieval is a property of the surface, not of the model. The same provider can reach your page through one product and answer from priors through… - [A page can render perfectly and be broken for machines](https://machinelayer.org/a-page-can-render-perfectly-and-be-broken-for-machines): The failure that is hardest to catch, because from where you sit the thing exists and is correct. - [An assistant can confidently describe outdated copy](https://machinelayer.org/an-assistant-can-confidently-describe-outdated-copy): The failure that looks most like success. The page was found, fetched and read correctly — just not this month's version of it. - [If it needs JavaScript, it arrives empty](https://machinelayer.org/if-it-needs-javascript-it-arrives-empty): A page that assembles itself in the browser is, to most automated fetches, an empty shell. This is most damaging where it is least expected: in the… - [Your sources may not be reachable](https://machinelayer.org/your-sources-may-not-be-reachable): Check that a citation can be fetched before you rely on it. The sources you reach for first are not always the ones that come back. - [Conclusion: Check what the crawlers get. The rest isn't yours.](https://machinelayer.org/check-what-the-crawlers-get-the-rest-isnt-yours): Being refused and being ignored produce exactly the same silence, and only one of them is your problem. Before concluding that a system will not read you… ### understanding — Having got it, did it parse what you meant? - Every model that reached the page parsed it correctly, every time. ### credence — Having parsed it, did it treat your claims as true? - [Put the source next to the claim](https://machinelayer.org/put-the-source-next-to-the-claim): A bibliography at the bottom is a gesture. A source sitting beside the sentence it supports is a resolution path. The difference shows up the moment… - [Publish the raw, not the conclusion](https://machinelayer.org/publish-the-raw-not-the-conclusion): Numbers presented as results ask to be believed. The inputs behind them ask to be recomputed, which is a much smaller request. - [A gap you leave gets filled by the reader](https://machinelayer.org/a-gap-you-leave-gets-filled-by-the-reader): Omitting a figure does not leave a blank. It delegates the figure. - [Say which claim you can't prove](https://machinelayer.org/say-which-claim-you-cant-prove): This is the highest-return sentence on any page, and almost nobody writes it. - [Name the thing you don't have](https://machinelayer.org/name-the-thing-you-dont-have): The same move pointed outward rather than inward, and the highest-returning sentence measured anywhere in this record. - [Predict the discrepancy before someone finds it](https://machinelayer.org/predict-the-discrepancy-before-someone-finds-it): If your number will not match what a reader computes, say so in advance and say why. This converts every future disagreement into a confirmation. - [Well-formed data reads as audited](https://machinelayer.org/well-formed-data-reads-as-audited): The uncomfortable one, and worth reading twice if you have just built something well-structured. - [Text aimed at AI reads as manipulation](https://machinelayer.org/text-aimed-at-ai-reads-as-manipulation): The most intuitive idea in this entire area, and the one that backfires hardest. If AI systems are reading your page, address them and tell them what… - [Conclusion: Aim to be auditable, not believed](https://machinelayer.org/aim-to-be-auditable-not-believed): Trust is a state in someone else's head that you cannot inspect. Auditability is a property of your page that you can. ### method — How to measure this without fooling yourself. - [Ask for a verdict, not a description](https://machinelayer.org/ask-for-a-verdict-not-a-description): The same model, the same page, the same day will tell you different things depending on the shape of the question. - [Test the consumer product as well as the API](https://machinelayer.org/test-the-consumer-product-as-well-as-the-api): A provider's developer API and its consumer chat product are different systems, and on the same day about the same site they answered differently: the API… - [A single check can pass on a stale copy](https://machinelayer.org/a-single-check-can-pass-on-a-stale-copy): A deploy is not atomic. Edge nodes take the new file at different times, so one passing fetch confirms one node at one moment, not a deploy. - [Use a canary phrase to date a read](https://machinelayer.org/use-a-canary-phrase-to-date-a-read): The hardest problem in this loop: models routinely answer from a cached copy of a page that no longer exists, and nothing in their answer says so. - [An improvement is not proof you caused it](https://machinelayer.org/an-improvement-is-not-proof-you-caused-it): Improvement is not attribution. If you iterate hard on one thing and the outcome improves, you will credit the thing you iterated on. - [Conclusion: Keep a record you can be wrong in](https://machinelayer.org/keep-a-record-you-can-be-wrong-in): The log is the instrument. Write down what you changed and why, including the changes you reverted — the why is what makes an old decision re-examinable… ### reference — The tags, fields and files the rest of this is about. - [](https://machinelayer.org/head): Every page, not only the home page — any of them may be the one that gets shared. - [The schema block](https://machinelayer.org/the-schema-block): One