Content freshness scanner
Paste one URL. A language model reads the page and lists the statistics, tool references, regulatory citations and time-bound phrases that have gone stale, in a suggested order of repair.
The analysis is model-generated. It is one model’s reading of one page at one moment, which makes it a starting point for a refresh rather than a finding. Caul is our measurement instrument and it is not involved here, so nothing this returns is a measurement, a share or a citation figure.
Enter a URL above. What comes back is one model’s reading of one page, produced in about a minute, and the limits below apply to every line of it.
Content goes stale in a small number of predictable ways, and the prompt asks for exactly those. Anything genuinely evergreen is meant to come back unflagged, so a short list is a normal result rather than a failed scan.
The P0 to P3 priority scale, and nothing else.
There is no composite, no grade and no number out of 100. Every flag gets one of four levels on the P0 to P3 scale below, and the tool reports how many landed at each. A count of flags is not a score of your page, and we do not convert it into one.
The scale is a rubric written into the prompt, not a calibrated instrument. The model applies it by judgement, the thresholds are ours rather than anyone’s standard, and two runs can place the same line differently.
Six things this tool is blind to.
A free tool that overstates itself costs more credibility than it earns. These limits hold on every run, and knowing them is the difference between using the output and believing it.
The disclosure for a heuristic.
- instrument
- A general-purpose language model, called through the Vercel AI Gateway. Not Caul, and not a measurement instrument.
- what was measured
- Nothing is measured. A model is asked to read one page and flag content it judges to be stale, against a written rubric.
- how
- The page is fetched once as served, parsed to text, headings, structured data and FAQ blocks, and passed to the model with the rubric. The reply is checked for shape, not for accuracy.
- over what window
- The moment you press the button. One page, one fetch, one run, no repeat.
- what this cannot tell you
- It assesses age and phrasing, never truth, so a flag is a prompt to check rather than a defect found. It cannot read content the raw HTML does not carry, cannot know about changes after the model's training cutoff, and cannot tell you anything about whether AI answers name or cite you. Two runs on the same page can differ.
For the question this cannot answer.
Whether AI answers name you, and which competitor gets named instead, is measured rather than assessed: a fixed prompt set, run repeatedly across the engines, with coverage, run count and window attached to every figure. The audit is free, delivered on a call, and you keep the report and the underlying data whatever you decide.