Scope and limits

What ShipJury reviews, and what it doesn't

A review is one automated visit to your public site, read by AI reviewers. It is deliberately narrow. This page says exactly where the edges are.

What the browser will and won't do

The browser (Playwright driving Chromium) opens your site in a fresh context with no cookies, saved logins or history, and tries to complete the goal you described using only safe, public interactions.

The browser will

  • Follow links
  • Open menus
  • Scroll
  • Click controls that don't submit anything
  • Visit public pages only

The browser won't

  • Type into any field
  • Create accounts or sign in
  • Submit forms or send messages
  • Buy anything
  • Download files

It stops at the point where an account, a message, a form submission or a payment would be required, and the report says that's where it stopped. If your goal can only be completed past that point, the review covers the path up to it.

Network limits

  • The site must be a live, public address served over HTTPS on the standard port.
  • Addresses on private or internal networks, local machines, and bare IP addresses are refused, both for the address you submit and for anything the page tries to load or link to.
  • Requests the browser refuses during a visit are listed in the report's trace, with the reason.
  • Pages behind a login, a paywall or a bot check are out of reach. The report will say what the browser saw instead.

The two viewports

ViewportSizeWhat it is
Desktop1366×850Desktop Chromium
Mobile viewport (emulated)390×780Desktop Chromium emulating a narrow touch screen with a mobile user agent — not a real phone

Viewport emulation catches layout and navigation problems that depend on screen width. It does not reproduce a real phone's browser engine, on-screen keyboard, performance or network. Problems specific to Safari on iPhone, for example, may not appear. Every mobile result in a report is labelled "Mobile viewport (emulated, 390×780)".

The review browser identifies itself with ShipJuryReview in its user-agent string.

How results are worded

Each viewport gets one result for the goal:

  • Verified The captured evidence shows the goal can be completed, up to the point where the browser has to stop.
  • Blocked The captured evidence shows something preventing it. This is only used when a screenshot or trace step shows the obstacle.
  • Could not verify The browser couldn't establish either way, for example because it ran out of steps or the capture was partial.

In a follow-up review, each original finding gets one status:

  • Still reproducible The same steps show the same problem.
  • No longer reproducible The same steps no longer show it. This is not proof that it is fixed for every visitor.
  • Could not recheck The steps couldn't be repeated, for example because the page changed too much.

What the reviewers are

Three AI reviewers read the same captured screenshots and page text — as a first-time visitor, a skeptical buyer and a mobile reviewer — and an editor merges duplicates, drops anything the evidence doesn't support, and keeps at most 3 fixes. No human reviewer is involved.

Every report also carries a critique of the whole site — content and wording, layout, visual design, offer and trust, and narrow screens — which is the reviewers' opinion, grounded in the screenshots and page text that were captured and marked on those screenshots where it can be. If you ask for a business-concept review, the report adds what the site says the business is, how clearly that comes across, what is convincing, what is risky and what it leaves unanswered; that is a reading of the idea as the site presents it, not market research, and it says nothing about demand or revenue.

Every report also has a simulated customer panel: eight people described by an AI, either from your choices (market, age, gender mix, kind of customer, comfort with technology) or picked from your site, each looking at the same screenshots on their own and saying what they understood, what held them back and what they would ask. They are AI role-play, not real customers or a survey, and the panel never predicts how many people will buy.

The site score is calculated from a fixed checklist of 25 yes/partly/no checks answered from the screenshots, and the report shows every check and its answer. It is a summary of one visit, not a prediction of demand, sales or conversion.

  • Findings are observations from one automated visit on one day. They can be incomplete or wrong.
  • What was observed is kept separate from guesses about the cause, which are labelled "Possible cause (unverified)".
  • A report with fewer than 3 findings, or none, is a valid result.
  • A review is not an accessibility audit, a security test, a performance test, or a legal or compliance review.
  • It is not a guarantee, and it makes no prediction about demand, revenue or conversion. Suggested fixes describe what a visitor should see; they don't know your codebase.

Limits on a single review

A review is bounded so it finishes in minutes and stays light on your site.

LimitValue
Total browser time for one review, both viewports together4 minutes
Time allowed for a single page to load25 seconds
Redirects followed for one page load8
Pages visited, per viewport6
Actions taken (clicks, scrolls, going back), per viewport8
Screenshots kept, per viewport10
Extra captures the editor may request to check a claim2
Interactive elements considered on one screen90
Visible text kept from one screen5,000 characters

If a limit is reached before the goal is settled, the capture is marked partial and the affected result is "Could not verify".

Privacy of the result

Reports and screenshots are stored privately and can be opened only by signing in to the account that ordered them. Review data is deleted after 90 days, or sooner if you delete it from the report. See the Privacy Policy.

Start a review — $19 USDSee the sample report