The problem with issue lists
Most automated audits produce dozens of findings. Without prioritization, these lists are paralyzing. You know something is wrong but not what to fix first, how long it will take, or whether fixing it will make a measurable difference.
What makes a recommendation useful
A useful recommendation answers five questions: what is the specific problem, why does it matter for this website, what exactly should be changed, how difficult is the change, and what improvement can be expected?
Generic advice — "optimize your images" — answers none of these. A specific recommendation — "the hero image on your homepage is 1.4 MB and is the primary LCP element; converting it to AVIF and compressing below 200 KB would reduce LCP from 4.2s to approximately 2.1s" — answers all of them.
Recommendation structure
Every AI recommendation should contain: the problem description, the current measured value, the target or expected value, the specific fix with implementation steps, the affected pages and the estimated impact level. Evidence — the actual data the recommendation is based on — should be attached.
AI Website Auditor
Ready to see what your website needs to fix?
Run a free audit to check:
How ordering works
Recommendations are ordered by a combination of severity, estimated user impact and implementation difficulty. A high-severity issue that affects every page and takes 30 minutes to fix ranks higher than a medium-severity issue that affects one page and requires a significant engineering effort.
The role of evidence
The evidence attached to a recommendation is what makes it trustworthy. Without evidence, a recommendation is an opinion. With evidence — the actual LCP value, the actual file size, the actual header response — the recommendation is grounded in what was measured and can be verified independently.