How to make your website agent-ready

That is the job. naitive.work finds the repeating work, scores it, and either names a tool you can try this week or takes the work on.

This week the loop is the public site itself: can an agent find it, read it, and use it. The live scan at /ready is how we check that. It is not the workflow audit, and it is not The Assessment.

The loop is the public site, not a new brand

Agent-ready is not a redesign. It is not a new logo. It is whether a fetch of the homepage gets a real page: HTTPS, HTML, a specific title and H1, a usable robots.txt, a sitemap, a contact or booking path, a real 404 on a missing URL, and the copy in the HTML rather than an empty app shell.

That is one workflow. Not "SEO." Not a department. Making the public site readable to an agent: fetch, score, fix the gaps, rescan.

Seventeen checks, three bands

The scan is our own catalog: 17 checks. Nine Essential, five Recommended, three Bonus. Essential carries most of the score. Four Recommended checks are scored every scan. A documented API or MCP is n/a on a brochure site. A shop is not failed for missing OpenAPI. Bonus (security headers, sitemap lastmod, language hints) cannot hide a broken Essential score.

The nine Essential checks:

• HTTPS and reachable. The live site responds over HTTPS.
• Real HTML body. A document comes back, not an empty placeholder.
• Title, H1, and meta description. Present, and not generic.
• robots.txt is present and usable. It exists, and it does not blanket-block the whole site.
• Sitemap is published. A working sitemap, linked from robots.txt.
• Contact or booking path is linked. One obvious path an agent can follow.
• Unknown paths return a real 404. Missing URLs return 404 or 410, not the homepage shell.
• Main content is in the HTML. A fetch sees the actual copy.
• Forms and buttons have accessible names. Where controls exist, they have a label.

The five Recommended checks:

• /llms.txt is present and useful. 80 or more words with links or headings is a pass. A thin marketing stub is partial. Scored every scan.
• Homepage can be read as markdown. Accept: text/markdown returns markdown, or a linked .md file. Scored every scan.
• JSON-LD names the business. Organization, LocalBusiness, ProfessionalService, LegalService, or RealEstateAgent is a pass. Other JSON-LD is partial. Scored every scan.
• Open Graph share metadata. og:title, og:description, and og:url. Scored every scan.
• Documented API or MCP surface. n/a on a brochure site. A shop is not failed for missing OpenAPI.

Bands: Not ready (0 to 49), Getting there (50 to 74), Agent-ready (75 to 100). The flavour of the copy changes by desk. The checks do not. Property, accounts, a clinic: Essential and Recommended are scored the same way.

Score this loop on the five signals

This is not a generic score. It is this loop, scored the same way as the free audit. The long form of the signals is in which workflows absorb well, and which don't. How to run the score on your own list is in how to audit your workflows for automation.

1. Repeatability

Strong. Same shape, different site: HTTPS, HTML, title, robots, sitemap, contact, 404, copy in the HTML, plus /llms.txt, a markdown homepage, and JSON-LD that names the business. A junior could follow a one-page list of what is still missing.

2. Already outsourced

Strong, and already separable. A junior, a developer, or a VA already owns the public site. Nobody treats robots.txt as the thing the firm sells.

3. Clear right answer

Strong. HTTPS, or not. Real HTML, or an empty shell. Title specific, or generic. robots.txt blocking, or not. Unknown path a 404, or the homepage again.

4. Measurable outcome

Strong. The live scan already measures it: 0 to 100, three bands. Agents can find, read, and use the public pages. Nobody is buying the fetch log.

5. Enough margin

Strong when the public site is how work arrives: the contact or booking path the scan already looks for. If the page is a brochure nobody follows, leave it.

Count how many signals score 4 or 5. Four or five: the loop is ready. Two or three: start narrow, one Essential check or a missing /llms.txt, markdown homepage, or JSON-LD. One or zero: you are looking at judgment, not a fetch. Leave it.

One tool this week

Do not buy a new website platform. Run the free scan at /ready, then put one small fix on the top gap: a failed Essential check, or a missing /llms.txt, markdown homepage, or JSON-LD that names the business.

Judge it by Friday. Rescan. The score moved, or it did not. If the fix is wrong, you learned that in days, not in a quarter.

Sometimes that is the whole fix. Sometimes the gaps are bigger than one check, and the next step on this loop is the €99 Agent-ready fix: implement the failed and partial findings, then rescan. That is a small opener on the public site. It is not The Assessment. You cannot tell which until this loop has a name and a score.

A person keeps the check

Software can fetch and score. It cannot decide what the firm offers, who it is for, or whether a page should stay private.

That check is judgment. Leave it on a person. The production around it, the HTTPS, the HTML, the title, the robots file, the sitemap, the 404, is what should come off the desk.

Common questions

Do this next

Score the loop that actually eats the week in five minutes at the free audit. It names the most absorbable workflow and ends with one tool to try this week.

If you also want to see what an agent can read on the public site, the live scan is at /ready. Extra, not the door.

If that pass finds something worth a wider map, The Assessment is the paid next step: €499, a 60-minute conversation, five to seven prescriptions across the business, credited 100% against any build.