The EntityMap validator tells you whether an entitymap.json file matches the EntityMap spec. It doesn’t tell you whether any AI system reads the file, or whether what the file says is true. I tested it on 9 October 2026.
What EntityMap is
EntityMap is a proposed standard for a file at the root of a site, /entitymap.json, with an HTML version at /entitymap.html. The file lists the entities a site covers, how they relate, and short passages from the site’s own pages as evidence. Each passage carries the publisher’s name, so attribution travels with it.
The v1.0 spec is dated 7 April 2026 and written by Fred Laurent and Dixon Jones. The site credits Waikay and InLinks Optimization Ltd, and Waikay sells the reference generator and a managed service for building the files. None of that makes the spec wrong. It’s worth knowing who wrote it.
How to use the validator
- Open entitymap.org/validate. There are two tabs: Paste JSON and Fetch from URL.
- Fetch from URL only works if the site serving the file allows cross-origin reads (CORS). If it fails, open the file in your browser, copy it and use Paste JSON.
- Press Validate. Results come in three groups: errors, which the page says you must fix before publishing; warnings, which are recommendations; and the checks that passed.
- Fix the errors, then read the warnings. Most are about optional fields, such as a Wikidata link for each concept.
The hard errors include a missing required field, a passage whose publisher name doesn’t exactly match the one at the top of the file, a passage over 600 characters, more than five passages for one entity, and an interpretive relation such as IMPROVES with no confidence value.
What I ran
I wrote a small test file with two entities from this site: the robots.txt snapshot method and the concept of blocking AI crawlers. Each passage is quoted from one of my posts. I didn’t publish the file. Then I broke it on purpose, one change at a time.
| Test | Validator result |
|---|---|
| My first draft | Invalid, 1 error. I had used MEASURES on a Methodology, and the spec only allows it on a Metric. A real mistake, caught. |
| Draft with that relation removed | Valid with 4 warnings: no Wikidata link for the publisher or the concept, and no relations between the two entities. |
| One passage credited to “jacobwhitmore.com” instead of “Jacob Whitmore” | Invalid. Publisher names must match exactly. |
| IMPROVES added with no confidence value | Flagged as an error. |
| Generated date set to 16 July 2026 | Valid, with a warning that the file is 84 days old and may be treated as stale. |
| One passage rewritten to say “robots.txt is a wall. Blocking a crawler keeps it out.”, pointing at a page that doesn’t exist on this site | Valid with warnings. |
The last row is the one to remember. The passage still carries the title of my robots.txt post but says the opposite, and its source link goes nowhere. The validator passed it, because it checks the shape of the file, not what the file says. The spec says each source URL must be publicly accessible. The validator didn’t check that either.
I also tried Fetch from URL on Waikay’s own published file. The validator couldn’t fetch it from my browser and suggested CORS or the paste tab.
What it can’t tell you
- Whether the content is true. See the last test above.
- Whether anything reads the file. The spec suggests an
EntityMap:line in robots.txt and labels it a proposed convention. RFC 9309, the robots.txt standard, doesn’t define it. EntityMap’s own case page rates its value for AI search engines today as emerging, not established. - Whether it’s certified. The spec says a certification registry launches in Q3 2026. On 9 October 2026, entitymap.org/certification returned a 404.
Who publishes one
On 9 October 2026 I requested /entitymap.json from the same ten UK newsbrands as my crawler study.
- Eight returned a 404: Sky News, the Guardian, the Independent, the BBC, the Mirror, the Sun, the Express and the Telegraph.
- The Daily Mail redirected to its homepage. No file.
- Yahoo (UK news): not checked. My browser couldn’t read the page, so I’m leaving it as a gap rather than guessing.
So none of the nine I could check publish one. Waikay’s own file was generated on 16 July 2026, 85 days before my check, which is past the 30-day point at which the spec says consumers may treat a file as stale. Waikay reports a before-and-after test on its own site. That’s one site, measured by the company that wrote the spec.
Should you use it?
If you’re going to publish an entitymap.json, run it through the validator first. It caught a real mistake in mine in seconds. Treat “valid” as the minimum: it means the file is well formed, not that it’s accurate, read or useful. Check each passage against the page it cites yourself.
I’m not publishing one on this site yet. What would change that is evidence I can check: an AI crawler requesting /entitymap.json in my server logs, or an AI company documenting that it reads the file. There are no new predictions in this post.