sounding.
Do your pages actually say anything?

A sounding is a depth check. Put your site in, and each published page comes back with a per-field verdict — keep, change, or too thin to cite — with a confidence score behind every call. Read-only. Nothing is rewritten, nothing is published.

What sounding does

1 · It reads your pages cold

Start from any URL. Sounding follows your links, reads the whole sitemap when there is one — up to 200 pages — and meets each page the way search and answer engines meet it: title tag, meta description, H1, FAQ content, and structured-data markup.

2 · It judges each field

Every field gets a typed verdict against fixed criteria — not vibes, not free-form chat. A confidence score comes with each call, so you know which verdicts to act on and which are too close to call.

3 · It tells you what to fix first

High-confidence recommendations rise to the top; low-confidence noise is dropped, not dressed up as advice. You keep full control: sounding reports, you decide, someone (or something) else edits.

The twenty-five fields, and what a verdict means

title
Scored from the raw fetched source, not from what a browser shows after script runs. keep when that source title is specific to this page. crawler-miss when it is the homepage title, a template title, or not the specific title in the source. The score says what the source title actually is. A more specific title helps only if the search engine receives that title.
meta description
keep when it summarizes this page and differs from siblings. change when it is boilerplate, vague, or mismatched to the content.
H1 heading
keep when it states what this page offers in its own words. change when it is a templated fill-in that reads identically everywhere.
FAQ
A special FAQ listing is scored only from markup in the raw source. Visible questions are not that listing. missing when the markup is absent. Do not infer it from questions on the page.
structured data
keep when schema.org markup is present and right for the page type. wrong when it misleads. missing when the page ships none — invisible to machines that could cite you.
unique vs siblings
Does this page say something its sibling pages don't? The recurring cause of "thin": one template, many URLs, no distinguishing substance.
citable
Could an answer engine — AI search — plausibly quote this page as a source for a specific question in its topic area? This is the GEO bottom line.
experience
keep when the page has specific concrete detail a template would not have. thin when it only makes generic claims of experience. missing when there is no experience signal.
expertise
keep when the page has specific subject substance. thin when it is slogans or filler. missing when there is no subject substance.
authoritativeness
keep when a clear responsible party, credential, or source identity is supported by the page itself. thin when a name appears with no reason to trust it. missing when there is no identifiable source.
trustworthiness
keep when claims are specific and not misleading. wrong when claims are misleading or unsupported. missing when there is no basis to trust the page.
OAI-SearchBot
keep when robots.txt does not disallow OAI-SearchBot. wrong when it does: the page is not shown in ChatGPT search answers, and blocking that bot keeps it out of summaries and snippets. A navigational link can still appear. Allowing the bot does not guarantee a citation. missing when robots.txt was blocked or not fetched. Do not guess.
GPTBot
Disallowing GPTBot is a training opt-out. It is not a search opt-out. missing when robots.txt was blocked or not fetched.
ChatGPT-User
ChatGPT-User is not the control for search appearance. wrong when robots.txt disallows it while leaving OAI-SearchBot allowed. missing when robots.txt was blocked or not fetched.
noindex
A noindex tag counts only if the crawler can read the page. wrong when the tag is present but the crawler cannot read the page. missing when the page was not fetched.
public HTML
keep when the fetch returned public HTML. wrong when the HTML is a login wall or share gate. missing when the fetch was blocked. Do not guess.
question in the H1
A page-shape check of the extracted H1, not a rule quoted from the crawler documents. keep when the H1 is a question. thin when that question is vague. missing when it is not a question.
direct answer
A page-shape check of the extracted text. keep when a direct answer is up top. thin when the opening is a slogan. missing when no answer was extracted.
named entities
A page-shape check of the extracted text. keep when specific names are written there. missing when none are.
dates
A page-shape check of the extracted text. keep when a specific date is written there. missing when none is.
comparison or source
A page-shape check of the extracted text. keep when a comparison or a source is written there. missing when neither is.
inbound mentions
missing unless the fetched text itself shows inbound mentions. Do not invent them.
off-site authority
missing unless the fetched text itself shows off-site authority. Do not invent it.
OpenAI listing
missing unless the fetched text itself shows a listing on an OpenAI site. Do not invent it.
search measurement
A field score is not evidence of more visits. missing unless the fetched text contains a before-and-after comparison of impressions, clicks, and average position. Do not invent numbers.

How to read the scores

A confidence score rides with every verdict:

A real result (billing-site homepage, first run): FAQ missing at 1.00, structured data missing at 1.00, H1 change at 0.72 — three same-day fixes, found in under a minute.

Questions people ask

What does "too thin to cite" mean?

It means an answer engine — AI search — would have no reason to quote the page, because it does not say anything specific. Usually that is templated copy that reads identically across sibling pages, a missing FAQ, or a page whose only substance is a headline. Sounding surfaces exactly which fields are thin so you know where the depth has to come from.

Does sounding edit or publish anything on my site?

No. Sounding is strictly read-only. It fetches your published pages, judges them, and returns verdicts. It never rewrites copy, never logs in, and never publishes a change. What you do with the verdicts is entirely up to you.

How are the verdicts produced?

Each page gets exactly one typed call to a judgment model (Typesafe System One, model jev-latest) against fixed, published criteria per field. The model returns a choice — keep, change, thin, missing, wrong — plus a confidence score. There is no free-form prose and no ghostwriting: the judgment is constrained to the criteria, which is what makes the confidence scores meaningful.

What sites can I check?

Any public http(s) site. Start from any URL: sounding follows the page's own links, and when a site renders with JavaScript and ships no links in its raw HTML, it reads the site's sitemap instead — up to 200 pages per run. As long as pages are publicly reachable without a login, they can be judged.

What does a run cost?

One judgment call per page. A quick scan judges the first 8 pages; a full scan judges everything the site's sitemap claims, up to 200 pages per run. No page content is stored: pages are fetched, judged in memory, and returned to your browser; the server keeps nothing.

Why is it called sounding?

A sounding is a depth measurement — dropping a line to find out how deep the water is. This tool drops a line into each of your pages and reports where the depth actually is, field by field, instead of guessing from the surface.