See what AI systems can actually access on your site.
Site Intelligence analyses the pages your site declares for discovery, checks how search and AI systems can access and interpret them, and turns technical and structural problems into a prioritised worklist.
Over 50 technical and AI-readiness checks, each with the evidence it was based on. A technical pass is not a promise of citation, and site health is one part of AI visibility: read the methodology.
What Site Intelligence analyses, and what it declines to judge.
An audit is only useful if its boundaries are stated. Four definitions govern every finding on this page.
Herm maps and analyses the pages declared through your site’s discovery infrastructure, and identifies important linked pages that fall outside them.
Over fifty technical and AI-readiness checks, grouped into access, extractability, structure, machine-readable signals and site architecture.
Every finding records what was observed, how many pages it affects, where in the site it originates, and the response or markup it was read from.
Not every check applies to every page. Herm marks a check as not applicable rather than penalising a site for something it could not fairly assess.
Not every AI crawler does the same job.
Herm separates access that can affect discovery and answer retrieval from access used for model training, so a deliberate training policy is not automatically labelled as an error.
| Purpose | Agents | State | What it means |
|---|---|---|---|
| Discovery and retrieval | Search and answer-retrieval crawlers | Allowed | Can affect whether pages are found and used in answers. |
| Model training | Training crawlers | Blocked | A policy decision. Herm describes the consequence and does not score it as a fault. |
| Unclassified | Agents with no declared purpose | Allowed | Worth an explicit decision rather than a default. |
See what disappears before JavaScript runs.
Your browser may see content an AI system never receives. Herm compares the information available in the initial page with the rendered experience to identify important content that depends on JavaScript, and reports which blocks were missing rather than a score.
Important content requires JavaScript. Herm reports which blocks were missing, not a score.
The checks that decide whether a page can be reached at all.
Status, canonical, redirects, indexability and discovery declaration. None of them are interesting on their own, and every one of them can silently remove a page from consideration.
What your pages state in a form a machine can use.
Structured data, metadata and declared relationships. Herm reports what is present, what is malformed and what is absent on the pages that matter.
Whether the page is organised into answerable parts.
Heading hierarchy, content blocks and internal linking decide whether a system can lift a passage that answers something, or only the page as a whole.
One template problem is one finding, not four hundred rows.
Findings are grouped by where they originate. A missing attribute on a product template affects every product page, and the work is one template change.
A prioritised list a developer can actually work from.
Every finding carries what was observed, how many pages it affects, where it originates, and the response or markup it was read from.
| Finding | Originates in | Pages | Severity | Layer |
|---|---|---|---|---|
| Important content requires JavaScript | Product template | 412 | High | Technical |
| Guides linked but never declared for discovery | Sitemap | 9 | High | Technical |
| Product structured data missing suitability attributes | Product template | 412 | Medium | Technical |
| Comparison page answers none of the tracked questions | /compare | 1 | Medium | Content |
| Redirect chain on legacy ingredient URLs | Redirect config | 38 | Low | Technical |
What to change, and where.
Each finding resolves to the layer that owns it: a template, a configuration, a discovery declaration or a piece of content.
Attributes exist but only after rendering. Fix the template.
Nine guides are linked but never declared. Fix the sitemap.
The comparison page loads perfectly and answers none of the questions buyers ask. That is not a template problem.
Site readiness and AI visibility belong in the same investigation.
Brand Visibility shows what AI systems say about the brand. Site Intelligence shows whether the brand’s own site is technically and structurally ready to support discovery and citation.
Which questions you appear in, how you are described, who appears alongside you, and which sources the answers cite.
Brand Visibility overview →Whether pages can be reached, whether the important information is in what a system receives, and what is getting in the way.
Together. Investigate both sides of the visibility problem without assuming correlation proves cause. Site Intelligence does not explain why a Brand Visibility score moved; it tells you whether the foundations were in place while it did.
Some fixes are technical. Others are content.
Site Intelligence can expose structural and content-quality issues. When the answer is better information rather than another technical change, Content Suite can help teams create or improve the content itself.
Attributes exist but only after rendering. Fix the template.
Nine guides are linked but never declared. Fix the sitemap.
The comparison page loads perfectly and answers none of the questions buyers ask. That is not a template problem.
Transparent evidence, protected implementation.
You can inspect every finding Herm reports and the artefact behind it. The check catalogue, weighting and calibration are Herm’s own work and stay inside the product.
- Every finding and the artefact behind it
- What was observed, and on how many pages
- Where in the site it originates
- The check catalogue
- Weighting
- Calibration
What an audit can tell you, and what it cannot.
Four limits worth stating before anyone builds a plan on this data.
Technical readiness removes barriers. It does not decide what an AI system chooses to use.
It is a policy decision with consequences Herm will describe, not a fault it will score you down for.
Structured data on a cited page is an observation. It is not the reason the page was cited.
What third parties publish about you, and what the answers already say, are the other parts.
Four teams, four reasons to open it.
The same audit answers four different questions, and each team reads a different part of it.
Find technical and structural barriers across the pages that matter, with scope figures that make the case for the work.
Understand whether the official site provides clean, accessible brand information, or whether third parties are the easier source.
Work from grouped findings with evidence attached instead of a scanner export with a row per URL.
Audit and prioritise client sites against one versioned methodology, and show the client the evidence behind every recommendation.
Common questions
Is blocking AI training crawlers a problem?
Not automatically. It is a policy decision with consequences Herm will describe, not a fault it scores you down for. Herm separates access that affects discovery and answer retrieval from access used for model training.
Will passing every check get us cited?
No. Technical readiness removes barriers. It does not decide what an AI system chooses to use, and site health is only one part of AI visibility.
Which pages do you analyse?
The pages your site declares for discovery, plus important linked pages that fall outside them. Where a check cannot legitimately run on a page, it is marked not applicable rather than counted as a failure.
Do we get a row per URL?
No. Findings are grouped by where they originate, so one template problem is one finding with a scope figure, not four hundred rows a developer has to deduplicate.
Does this explain a change in our visibility score?
No, and it deliberately does not try. It tells you whether the foundations were in place while the score moved. Correlation between the two is not evidence of cause.
Find out what AI systems actually receive from your site.
One audit across the pages you declare for discovery, with the evidence behind every finding and a worklist grouped by what would fix it.
Sample figures throughout.