Landing page audit (heading hierarchy, CTA and an optional LLM review of the hook)
A landing page audit in Web Check shows how a page is built for the person who reads it: whether the headings follow a clear order, whether paragraphs turn into walls of text, whether a call to action is there. The fast layer is a heuristic that needs no neural network and answers in seconds. The deep layer is a model review that you switch on with one flag.
What a landing page audit looks at
A landing page audit starts with the engagement section, which reads the page structure by rules, without a neural network. It looks at:
- the hierarchy of headings from H1 to H6: no H1, several H1, skipped levels;
- walls of text, meaning paragraphs that run too long;
- lists, which break a text into scannable parts;
- the length of headings, where one longer than 60 characters becomes a remark;
- calls to action (CTA);
- readability of the text.
Each finding comes with an importance level, so a missing H1 and a long heading do not look alike in the report.
Two layers: a quick heuristic and a deep model review
A landing page audit has two layers. The heuristic layer works on every call and takes seconds. The deep layer starts only when you add the &llm=1 flag. Then a language model evaluates four things: the hook of the first screen, the "benefit against feature" balance in headlines, how convincing the CTA is, and the structure of the text.
The answer is a list of "hooks". Every item names its scope (hook, headline, cta or structure), its importance, an explanation and a ready rewrite that you can paste. The prompt rests on conversion research (Baymard, Cialdini, Nielsen) and on direct marketing (Ogilvy, Schwartz, Sugarman).
How an audit runs, step by step
- You enter the page address, or call
GET /api/engagementwith theurlparameter. - The heuristic reads headings, paragraphs, lists, CTA and readability and returns an engagement score within seconds.
- With
&llm=1the model reviews the page and returns the list of hooks with ready rewrites. - The final engagement score is the lower of the heuristic score and the model score. The model cannot lift the grade above what the heuristic gave.
- A repeated model review for the same address on the same day comes from the cache.
What you get from the audit
- An engagement score and a list of remarks on structure, headings and CTA.
- With the deep layer, a list of hooks, each with scope, importance, explanation and a rewrite that is ready to use.
- The same data as JSON for your own tools.
A landing page audit of this kind shows how the page is organised and where the wording could be sharper.
Where it helps and how to run it
- Before a campaign. Run a landing page audit to check the first screen and the CTA of the page you are about to send traffic to.
- After a redesign. Check the heading hierarchy for a second H1 or skipped levels left behind after blocks were moved.
- When texts are rewritten. The ready rewrites from the model give you a starting point for the next draft.
GET https://webcheck.asibots.pro/api/engagement?url=PAGE_URL
GET https://webcheck.asibots.pro/api/engagement?url=PAGE_URL&llm=1The heuristic is also part of the full audit and of the nightly auto-audit of ASI Bots' own sites on Monday, Wednesday and Friday. Keyword placement is a different question and is covered by page-audit.
Honest gates and limits
- On a protection stub, an error page or an empty JavaScript shell, the model review does not start. Instead of advice for an error page, Web Check returns the reason.
- One model review per address per day: a repeat comes from the cache, which saves requests.
- The model cannot raise the score above the heuristic one, so a high grade means both layers scored the page high.
- The deep layer rests on the principles of the cited research and does not replace testing on real visitors.
Frequently asked questions
Can the model inflate the engagement score?
No. The final score is the lower of the heuristic and the model scores, so the neural network cannot push it above what the rule-based layer gave.
What happens if my landing page is hidden behind a protection page?
The model review does not start. On a protection stub, an error page or an empty JavaScript shell Web Check returns the reason instead of advice.
How often can I run the deep landing page audit?
One model review per address per day. A repeated request is answered from the cache. The heuristic part needs no model and takes seconds.