Client project • Name withheld • AI-readiness audit
Can AI search tools and AI agents actually read and understand this website? For a long-established Indonesian B2B company, the answer was only partly. The site scored 52 out of 100, and the reasons were not the usual suspects: crawlers were welcome and the titles were good, but a CAPTCHA, the wrong schema language and almost no semantic structure held it back.
Project snapshot
The site is a WordPress company profile written in Indonesian and served as static HTML. The audit combined an automated AI-readiness scan (WordLift AI Audit) with my review of what each finding means for this business and which fixes should come first.
The client's name, URLs, file names and identifying figures from the original report are withheld to protect the client's privacy.
Starting point
Three areas scored poor, four needed improvement, and three were already good.
| Area | What the audit found | Rating |
|---|---|---|
| Automation readiness | A CAPTCHA on the page blocks AI agents from using the form (priority 1) | Poor |
| HTML semantics | Semantic ratio of 2%: content sits in generic containers with little heading structure | Poor |
| Content structure | Very little readable main content, not broken into clear sections | Poor |
| Structured data | Four microdata types; schema declares English while the content is Indonesian; Organization missing address, contact point and social profiles | Needs improvement |
| Site files | All 23 AI crawlers allowed and llms.txt present; no skill.md or agent-skills index | Needs improvement |
| Content freshness | An experience figure and the footer copyright year were out of date | Needs improvement |
| Image accessibility | Three images without alt text, including the logo | Needs improvement |
| SEO fundamentals | Clear, descriptive title and meta description | Good |
| Internal linking | Strong internal link structure | Good |
| JavaScript rendering | Static HTML, so all content is visible to crawlers | Good |
Key findings
CAPTCHAs stop bots by design, including AI agents that try to send an enquiry for a user. The fix is lighter protection, such as a hidden honeypot field or a managed, mostly invisible challenge, which still filters spam.
Structured data declared the content as en-US. Every entity should state id-ID, so search engines and AI systems interpret the language, and the audience, correctly.
With a 2% semantic ratio, machines could not tell the main content from the layout. Article and section elements, real H2 and H3 headings, and question-based sections fix that.
The Organization entity had no address, contact point or social profiles, and the WebSite description was empty. Moving everything to one JSON-LD graph makes the company's facts explicit and easy to maintain.
An outdated experience figure and last year's copyright line look minor, but AI systems and buyers both read them as signs of an unattended site. Three images also needed alt text.
Recommended fix order
| Phase | Work | Effort |
|---|---|---|
| 1. Quick wins | Set schema language to id-ID, add alt text to three images, update the experience figure and copyright year | Low |
| 2. Entity and schema | One JSON-LD graph with Organization address, contact point and social profiles, plus a WebSite description; retire the microdata | Low to medium |
| 3. Structure | Semantic HTML, a clear heading hierarchy, question-based sections and fuller main content | Medium |
| 4. Agent access | Replace the CAPTCHA with lighter bot protection; optionally add skill.md and an agent-skills index | Medium |
Measured baseline
Results are added here only after the changes are live and re-scanned. Until then, this page shows the diagnosis, not an outcome.
| Metric | Baseline, September 2026 | Target |
|---|---|---|
| AI-readiness score | 52/100 | Re-scan after phases 1 to 3 |
| Semantic ratio | 2% | Clear main-content structure |
| CAPTCHA on the enquiry form | Present, priority 1 | Replaced with lighter protection |
| Schema language | en-US | id-ID |
| Images without alt text | 3 | 0 |
| AI crawlers allowed | 23 of 23 | Keep |
Questions
Because buyers increasingly ask AI tools which suppliers to consider. A company-profile site is often the only source those tools can read about the business, so if it is hard to parse, the business is hard to recommend.
Not directly, because search crawlers do not submit forms. It matters for AI agents that try to complete a task for a user, such as sending an enquiry. A lighter form of bot protection keeps spam out without blocking them.
To show how the diagnosis works on a real site. Results are only added once changes are live and re-measured. Publishing numbers before that would be guessing.
Send your website. I will run the same audit, review what each finding means for your business, and send a prioritised fix order.
Last updated