What a Go Voltic report measures, and how it says so.
Last updated 2026-10-20 ยท describes the report format currently shipping
Every line in a Go Voltic report carries a label that says how we know it. This page is the contract behind those labels: which instrument produced each kind of claim, what is read in full, what is sampled, what is never checked, and how a refusal is reported. If a report and this page ever disagree, tell us at hello@go-voltic.com and we will fix whichever one is wrong.
Two instruments, named on every claim
The free scan reads what your public pages serve. It fetches your pages the way a fast, script-less visitor would: the markup, the headers, the structured data, the response timing from our edge. It presses nothing, submits nothing, and signs in to nothing.
The Deep Scan opens your site in a real browser. It loads and times up to 40 of your public pages, presses safe controls on a sample of up to three of those pages while recording what visibly changed, and runs an automated accessibility battery on a sample of up to eight rendered pages. On up to four pages it then does what a code reader cannot: it presses Tab through the page, loads it fresh at 320 and 390 pixels wide, and enlarges the text spacing, taking a screenshot where something breaks. Findings from the walk say what was pressed and what happened; they are the record of an instrument, not a person's impression.
Some readings are Google’s, and say so. Google’s PageSpeed Insights runs its own lab test on your home page and up to two other templates, and Google’s Chrome UX Report supplies real-visitor speed only where Google publishes it for your site. The report prints both as Google’s figures, beside our own timings, never blended into them.
The two instruments disagree sometimes, and the report says which one is speaking. A page that serves clean markup can still have a dead button; only the walk finds that. A page that renders beautifully can be unreadable to a crawler; only the read shows it.
The evidence labels
| Label | What it means |
|---|---|
| Measured | The scanner recorded a timing, a response, or a test result. The number in the report is the number the instrument produced. |
| Observed | The scanner saw the element, the header, or the destination itself: it exists and this is what it contained. |
| Inferred | The evidence points somewhere, but the scan did not measure the business outcome. Inferred lines say so, and dollar figures attached to them are always ranges, never promises. |
| Not checked | The scope could not establish it. It never counts as a pass, and the report says it rather than leaving the row out. |
The last row is the one that separates an honest report from a flattering one. A report that quietly omits what it could not test reads exactly like a report that found nothing wrong. Ours prints the gap.
Scope, stated as numbers
- Up to 40 public pages are opened and timed per Deep Scan.
- Safe controls are pressed on a sample of up to three pages; the accessibility battery runs on a sample of up to eight rendered pages.
- The keyboard, phone-width and text-spacing walk covers up to four pages, up to 50 Tab stops each, with up to six screenshots as evidence.
- Every file the opened pages load is checked for errors, redirects on the way in are counted, and up to 60 further links are checked at one request a second. Up to five PDFs linked from your own pages are read for tags, language, title and text.
- Privacy is tested on three fresh visits to your home page: a plain arrival, one that presses the cookie banner’s reject button where there is one, and one with Global Privacy Control on. We press only a reject control: never accept, never a form’s submit button.
- Google’s lab test runs on your home page and up to two other templates, on Google’s mobile settings. Lab figures vary by a few points run to run, and they move no score.
- Pulse also reads your home page once a day between monthly walks. It calls the page down only after two failed checks in a row, and a bot check or a refusal is never reported as down.
- Every Deep Scan tests the site against a 21-standard agent-readiness catalog: the schema, feed and machine-navigation standards an AI shopping assistant looks for, each printed as met, not met, or not measured.
- Real-visitor speed appears only where Google publishes it (Chrome UX Report p75) for your origin. Where it does not, the report says there is no field data instead of padding the page.
- Every report states how many addresses were found, attempted, and read, as counts. When the true size of a site is not knowable from outside, we do not print a percentage that pretends it is.
- There is no human visual-design review in the self-serve products (screenshots appear only as evidence for a finding), no security or penetration testing, no pages behind a login, and no real-visitor analytics outside the consent-gated behavioral tier.
When a site refuses us, we say who refused
The scanner honors robots.txt before it reads anything else, identifies itself honestly on every request, and treats a bot wall as a full stop, never something to get around. When access is refused, the report attributes the refusal: the robots policy, a named security vendor, or the origin itself. A refusal aimed at our scanner is a fact about our access, not about whether your site works, and we never report it as "your site is down." The crawler documents itself at go-voltic.com/crawler, including how to keep it off your site entirely.
How the score is built
Five areas are scored, and the report names them the way it prints them: structure and content, machine-readable foundations, speed signals, accessibility, and the trust and interaction checks measured on the pages we pressed. Privacy is the sixth system we read, and deliberately not a scored one: the report flags exactly what loaded before consent, on which pages, as a reading with evidence, and never averages a legal-risk flag into the number. An area with no evidence on a given scan is excluded from the average entirely rather than given a middling score, and the report says so beside the number.
Since August 2026 a finding can also say how common the issue is: where at least 30 comparable measured scans exist for that exact check, the report prints the share of measured sites carrying it, with the sample size, a 95% interval, and the window it was measured in. The comparison set is frozen into your scan when it runs, so the sentence can be re-derived from your document alone. A check measured on fewer than 30 scans prints no share at all, and a check younger than its window counts only the scans that actually ran it. Below the floor we stay silent rather than estimate.
Whether an AI answer engine can read and quote your pages is part of the score: structure, schema, crawler access, and answer shape, measured on your pages. Whether an engine actually cites you against real buyer questions is a different measurement, and it is the one the AI Visibility Report makes.
The two AI reports each print a score of their own, built differently and named differently so neither is mistaken for this one. The AI Visibility Score is 40% technical eligibility (the average of crawler access, readability without JavaScript, structured data and entity clarity) plus 60% observed citation (the share of Claude’s answers, with live web search, that cited you), so a site no engine cites cannot score above 40. The Head-to-Head score is the share of usable answers that named or cited a brand, and every brand in the comparison is scored the same way. Both reports also put the same questions to Perplexity, which searches its own index, and print its reading beside Claude’s. Perplexity never enters either score, so a score keeps its meaning from month to month. Each score carries a version, and a score is only compared with one of the same version. A report whose score could not be measured is not sent; we run it again.
The free scan prints a score of its own, from a lighter read of fewer pages. A paid report’s number will not match it, and the difference can run either way.
How the findings and benchmark figures are counted
The findings pages and the benchmarks print counts from our own scanner. They are not a survey and not an estimate. A finding’s share is the number of free four-page scans that raised it, divided by every free scan in the stated window that loaded at least one page. Those scans include Go Voltic’s own research sweep over a pool of small-business sites as well as scans run by visitors, and each page says how many of each.
Deep scans read a site at a different depth, so they are a separate population and are never in a free scan’s denominator; the sector benchmark pages print the deep population on its own. A site is counted once in the distinct-sites figure, however many times it was scanned. Every share carries its sample size and a 95% Wilson interval, because a share printed without them reads as certainty and is indistinguishable from a guess.
What this methodology will not claim
No automated scan can certify legal, accessibility, or privacy compliance, and we do not. No honest audit can guarantee rankings, traffic, citations, or revenue, and we do not. Dollar estimates in any report are hedged ranges built from stated assumptions, labeled Inferred, and never guaranteed.