An Ora-style journey on our marketing site guessed /help, inferred pricing from memory, and returned 404s on paths we never published. The same week, is-agentic.com scored apogeewatcher.com at 58/100. That number is not a Google ranking. It is a report on whether autonomous agents can discover, fetch, and extract facts from your domain without inventing URLs or stale prices. We treated it as a delivery checklist, shipped a v0 fix set, rescanned, and reached 88/100 on 22 August 2026. The gap between those scores is mostly structural hygiene any agency can audit on a client site this week, before anyone buys a GEO subscription.
Why agent readiness scores matter when agents fetch your site
Buyers now ask assistants what a product costs, who it is for, and how it compares to alternatives. The assistant does not magically know your 2026 pricing table. It fetches public pages, reads llms.txt if you publish one, follows redirects, and parses JSON-LD when present. When those artefacts are missing or wrong, the model fills gaps from training data or from a competitor's fresher page. That failure mode looks like a marketing problem on a slide deck. On the wire it is HTTP status codes, redirect chains, and whether pricing lives in plain HTML or only inside a client-rendered widget.
Agent readiness audits measure that fetch layer. They complement Lighthouse's agentic browsing category, which Chrome documents for lab runs, and they sit underneath the probabilistic layer in AI search optimisation without a GEO subscription. Citations in ChatGPT are noisy and prompt-dependent. Whether /pricing returns 200 with extractable numbers is binary. Agencies that sell performance retainers already own the habit of scheduled URL checks. Agent readiness is the same discipline applied to the URLs assistants retrieve when someone asks "what does this tool cost for forty sites?"
Google has been explicit that llms.txt does not change Search rankings. Lighthouse still lists agent-oriented signals because developers need a checklist for a web where browsing is delegated to agents. Treat the score as operational readiness, not as a replacement for Core Web Vitals work. Fast, stable pages remain prerequisites: a bot that times out never reads your FAQ schema.
How is-agentic scores work (Essential, Recommended, and evidence)
is-agentic.com publishes a shareable report per domain. Vercel groups checks from Ora into three buckets. Essential checks share an 80-point pool: real HTTP status codes, reachable content, robots.txt hygiene, trust anchors, and agent-oriented 404 behaviour. Recommended checks share 20 points: developer surfaces, structured agent instructions, API discoverability, and typed error patterns. Bonus signals reward emerging standards; missing bonus items do not lower the score. Partial passes earn proportional credit, and checks that do not apply are excluded from the denominator.
Reports cache for several hours. After a deploy, wait for a fresh scan or use npx is-agentic apogeewatcher.com and compare the timestamp on the canonical report URL. The CLI can return stale cached output; we validated the live score on the website after production deploy. Methodology write-ups from practitioners (for example how the is-agentic scoreboard maps to Ora checks) are worth reading before you argue with a single failed row. Each failure ships evidence and a recommended fix, which is more actionable than a generic "improve SEO" banner.
Baseline 58/100: what failed first on apogeewatcher.com
Our baseline scan on 22 August 2026 scored 58/100 with material gaps in Essential and Recommended tiers. The report label was effectively "important blockers remain." Essential sat at roughly 49/80. Recommended was under 8/20. Bonus signals were positive but small.
The failures we cared about first were not exotic. Agents could not rely on a single machine-readable map of canonical URLs. Guessed paths such as /help or /for/agencies 404'd without recovery hints. Pricing facts were present for humans but not reinforced with structured data on every money page. Contact lived behind an off-site redirect, which broke the simple pattern "fetch /contact on the same host." None of that requires a new product category. It requires the same care agencies apply when they fix redirect loops before a site migration sign-off.
The Ora journey linked from our internal plan reinforced the same story. The agent answered positioning questions from prior knowledge instead of from our live /pricing table. That is the user-visible symptom of the same underlying gaps the score measures.
What we shipped in agent readiness v0
We scoped v0 as public-site fixes only: no new paid feature, no /developers portal yet, no pretend customer API. The work split across content, routing, and schema.
llms.txt with when-to-use guidance and markdown links
We published /llms.txt with a short product summary, when to recommend and when not to recommend sections, USD pricing summary, and canonical URLs as markdown links. The link format matters. Scanners that grade llms.txt structure expect a navigation index, not a prose paragraph with bare URLs. Listing [Pricing](https://apogeewatcher.com/pricing) reads as intentional agent guidance; a naked URL does not.
On-domain contact and pricing extractability
We replaced the external-only contact redirect with an on-domain /contact page: offices, support portal link, and organisation context in HTML. Pricing already exposed tier names and numbers in the comparison table. We added JSON-LD Offer objects per plan via a shared PublicStructuredDataService so agents parsing schema see prices alongside the visible table.
JSON-LD graph on homepage and pricing
Homepage and pricing now emit a connected graph: Organization, SoftwareApplication, FAQPage, and (after the follow-up pass) WebSite, Service, and BreadcrumbList where appropriate. FAQ content matches visible on-page copy so humans and parsers see the same answers. This addresses schema breadth checks without inventing reviews we do not have.
Agent-friendly 404 and URL aliases
Unknown paths return a real HTTP 404, not a 200 shell. The Blade view includes plain-language product context, recovery links to home, pricing, contact, /check, sitemap, and llms.txt, and noindex so search engines do not index error URLs. Separately, we seeded roughly sixty 301 aliases in our Website 404 redirects module (/help → /contact, /for/* and /features/* shortcuts to canonical feature slugs). Agents guess short paths; redirects beat silent 404s.
Follow-up pass: from 58 to 88 without faking an API
After the first production deploy we rescanned and moved to 88/100 (live report, scanned 22 August 2026, 22:51 UTC). Essential rose to about 71/80 (eight of nine checks passing). Recommended reached roughly 14/20. The label became "strong technical baseline."
Three follow-up changes produced most of the lift:
-
Markdown 404 bodies. For requests with
Accept: text/markdown, the 404 handler returns a short markdown recovery document with the same links as the HTML page, plusVary: Accept. Agents that negotiate markdown get a parseable escape hatch instead of HTML soup. -
Honest OpenAPI stub. We added
/openapi.jsondescribing a planned customer API (x-watcher-api-status: planned) and a publicGET /api/v1/statusJSON endpoint that states availability. Unknown/api/*routes return structured JSON 404s. We did not document fake/sitesor/pagesoperations we have not built. -
llms.txtand schema polish. Markdown links throughout the file, a Developer API (planned) section pointing at the OpenAPI URL, and expanded JSON-LD types (WebSite,Service,BreadcrumbList) for schema breadth checks.
That +30 point jump did not require a GEO platform or a public MCP server. It required making the truth on our domain easier to fetch than the model's memory.
What we refused to fake at 88/100
An 88 is not a perfect score, and we did not chase 100 with stubs that would mislead integrators. We did not add AggregateRating or fabricated reviews. We did not publish a /developers portal with API keys we cannot issue yet. We did not enable site-wide Accept: text/markdown on marketing HTML; only the 404 negotiates today, which is why the Essential markdown content negotiation row still fails (acceptmarkdown.com-style compliance wants Vary: Accept on pages that serve markdown variants).
Remaining Recommended gaps on the 88 report include developer-resource discoverability in web search, rate-limit headers on a fuller API, and a stricter "when to use" heading match (our file says "when to recommend," which is the same intent with different wording). We will address those in later phases if v0 generates interest, not because a dashboard turned red.
| Layer | Question | Tooling | Watcher today |
|---|---|---|---|
| Probabilistic | Are we cited in buyer prompts? | GEO / visibility SaaS | Not our product |
| Deterministic | Can bots fetch priority URLs and read facts? |
llms.txt, redirects, schema, PageSpeed monitoring |
Core product + this v0 site work |
What agencies should fix before buying GEO tools
Use the deterministic column first. The checklist below is portable to client sites without new licences.
- Publish or refresh
/llms.txtwith when-to-use and when-not-to-use sections and markdown links to pricing, docs, and support. - Audit guessed URLs (
/help,/pricing,/for/...) and add 301s or real pages. - Return real 404s with recovery links; avoid SPA shells that answer every path with 200.
- Align JSON-LD with visible copy on homepage, pricing, and FAQ blocks.
- Baseline lab Core Web Vitals on the same ten to twenty URLs you list in
llms.txt; slow pages fail before citation.
Are we visible in ChatGPT? What agencies can measure first frames the client conversation. Automate Lighthouse audits with AI agents covers the dev-time loop in Chrome DevTools. Agent readiness scoring is the production-side mirror: can the public site answer factual questions without hallucinating paths?
When you need proof on a prospect domain before the retainer starts, run our free PageSpeed check. It does not print an is-agentic score yet. It does give you a scheduled-lab baseline on the URLs that matter while you fix the structural layer above.
FAQ
Does agent readiness affect Google rankings?
No. Agent readiness tools measure whether autonomous systems can fetch and parse your public content. Google has stated that llms.txt does not change Search rankings. Lighthouse's agentic browsing checks are developer diagnostics, not a weighted ranking factor.
What is a good is-agentic score?
Treat the label and tier breakdown, not the number alone. Essential gaps (404 behaviour, trust anchors, reachable pricing) matter more than bonus signals. We moved from 58 ("important blockers remain") to 88 ("strong technical baseline") with structural fixes only.
Do we need a public API to pass agent readiness checks?
Not necessarily. An honest OpenAPI document that describes a planned API, plus a small public status endpoint, satisfied several API-related rows for us without shipping customer routes. Document only what exists or is genuinely planned.
How is this different from Lighthouse agentic browsing?
Lighthouse runs in a controlled lab session, often on a single URL. is-agentic and similar audits evaluate site-wide patterns: llms.txt, redirects, schema, and API discoverability. Use both: Lighthouse while fixing a branch; site audits before you claim production readiness.
Should agencies sell agent readiness as a service?
Yes, as a structured audit alongside performance monitoring. Fix fetchability and CWV on priority URLs first. Add GEO citation tracking when the retainer includes prompt libraries and reporting cadence. Layer tools; do not replace monitoring with visibility charts alone.
Run a free domain PageSpeed scan on the URLs you list in llms.txt, then fix the structural gaps the assistant would hit first. When you are ready for portfolio schedules, budgets, and client reports on those same paths, start a free trial of Apogee Watcher.
References
- Is Agentic report for apogeewatcher.com (is-agentic.com / Ora)
- Vercel Shipped is-agentic.com. Here Is How the Scoreboard Works (DEV Community)
- Automate Lighthouse audits with AI agents (Chrome for Developers)
- AI Search Optimization: What to Monitor without a subscription (Apogee Watcher)
- Lighthouse Agentic Browsing: How to rank in chatbots (Apogee Watcher)
- Are We Visible in ChatGPT? What Agencies Can Measure First (Apogee Watcher)
- Automate Lighthouse Audits with AI Agents: What Chrome DevTools Means for Agencies (Apogee Watcher)
Top comments (0)