Documentation
Everything you need to run your first scan, test a real flow and act on the report: Instant Scan for any public page, the Chrome extension for signed-in flows, the fix prompt for your AI editor and the MCP server.
Overview
Walkthru shows you where a first-time visitor would get stuck on your site. There are two ways to use it.
- Instant Scan
- Paste an address and get a report in about 20 seconds. No account and no install. It reads your public pages and checks first impression, SEO basics and passive security headers.
- Test runs
- An AI test user works through a real goal, like signing up, in your own browser tab. The Chrome extension clicks and types; our server decides every step and notes where the test user got confused.
Both end in the same report: what happened, the evidence, and the fixes worth doing first.
Run an Instant Scan
- Go to the scan form on the home page, or the Instant Scan panel in your dashboard.
- Enter your address, like
yoursite.com. - Select Scan my site. The report opens when it is ready.
The scan follows up to 10 same-site links from your homepage, honours robots.txt and only sends ordinary page requests. Pages that are built entirely by JavaScript are marked as checked "before JavaScript runs", so you know what was and was not measured.
Each network address can run five Instant Scans an hour. Scan reports get a public link so you can share them.
Install the extension
Test runs need the Walkthru Chrome extension. During the beta you load it yourself: download the zip you were sent, unzip it, open chrome://extensions, turn on Developer mode and choose Load unpacked.
- Add Walkthru to Chrome and pin the bird to your toolbar.
- Sign in to your dashboard with Google, GitHub or an email link.
- On the dashboard, select Connect extension. This hands your session to the extension so it can start runs as you. Nothing else receives it.
The extension only acts on the tab you start a test in. See extension permissions for what each permission is used for.
Run a test
- Open the site you want tested in a normal Chrome tab. If the goal needs an account, sign in to your site first.
- Select the bird in the toolbar. The side panel opens beside the page.
- Write a goal in plain words and pick a test user.
- Tick This is a logged-in page when testing a dashboard or account area. The test user will ask you before submitting any form there.
- Select Start test. The first time, Chrome asks for permission to capture screenshots of the tested tab.
You will see each step and the test user's thoughts in the panel, and Scout, the Walkthru bird, in the corner of the page. A free run stops after 12 steps; paid plans allow up to 30. Runs also stop when the goal is reached, after 4 minutes, or if the test user leaves your site. Select Stop at any time; the steps so far still become a report.
Choose a test user
Each test user behaves like a real kind of visitor, so they get stuck in different places. Try more than one on the same goal.
- First-time visitor
- Has never heard of your product, skims and gets impatient fast. Catches unclear headlines and too many steps.
- Phone user
- On a small screen, taps big obvious things and hates long forms. Catches cramped layouts and tiny targets.
- Small-business buyer
- Deciding whether to pay. Looks for pricing, trust and what happens next. Catches hidden costs and missing proof.
- Skeptical developer
- A cautious developer who reads error messages, checks links and distrusts vague copy. Catches silent failures.
On the Plus plan you can also describe your own test users in Settings (up to 10), and tick several in the extension. They run one after another on the same goal, each starting on the same page, and every report of the set shows them side by side with the problems more than one of them hit. Each test user uses one run.
Write a good goal
A goal is what a real visitor came to do. Keep it to one outcome the test user can recognise when it happens.
- Good: "Sign up for a free account", "Find the price of the Pro plan", "Send a message through the contact form".
- Too vague: "Test the site", "Check everything".
- Too many goals: "Sign up, create a project, invite a teammate and upgrade". Split this into separate runs.
Safe mode
Safe mode is always on. It is enforced in code, not left to the AI.
- Never clicks buttons that pay, delete or cancel.
- Never opens
mailto:ortel:links. It records them as working contact methods. - On logged-in pages, asks you in the side panel before submitting any form.
- Buttons that send or invite only run on a verified domain, after you approve in the side panel, and at most once per run. Everywhere else the run ends as "Stopped before sending".
- Stops at a CAPTCHA. It never tries to solve one.
For dashboard tests we still recommend a test account with sample data.
Why can't Walkthru test Instagram?
Test users run in your own browser, with your own logins. On sites you do not own, a test user that likes, follows, posts or buys would act as you, could get your account banned, and would test someone else's site without their permission. So Walkthru refuses banking, trading, social, webmail and similar sites outright, and on any site you have not verified it acts as a visitor.
- A visitor reads, scrolls, clicks links, menus and tabs, and uses the site's search box. It never fills in other forms, signs in, likes, follows, posts, messages, buys or adds to a cart, and it stops if the page shows you are signed in.
- On your verified domain it does what a real visitor does, including sign-up forms and signed-in pages, with safe mode still blocking payments, deletions and cancellations.
These rules are checked by our server at every step, not by the AI, so no goal or page text can talk a test user past them. If a site shows a CAPTCHA or bot check, you solve it yourself and press Continue, or the report says the site's bot protection stopped the test.
Verify your domain
Verifying proves you own a site. It unlocks the full security check (exposed files and backups, keys leaked into JavaScript, public source maps and subdomain takeover) and lets approved send actions run during tests.
- Open Settings and enter your website address.
- Recommended for live sites: add the displayed TXT record at your DNS provider. No source changes or redeploy needed. For example.com the record is
_walkthru.example.com; for app.example.com it is_walkthru.app.example.com. Some providers append the domain automatically, so check the final name. - Press Check verification. If DNS has not updated yet, wait for your provider to publish it and try again. Ownership is checked again when a run starts.
<meta name="walkthru-verification" content="your-token">If you cannot manage DNS, open Other ways to verify for the homepage tag above or a file at /.well-known/walkthru.txt. A hosted subdomain may need its administrator to publish proof. A confirmation checkbox alone does not prove ownership.
Public competitor comparisons do not need verification. Use Compare for public metrics, or browse signed out with a visitor goal in the extension.
Read a report
Every report leads with a summary and three numbers: findings by severity, the outcome, and the peak confusion (0 to 3) with the step where it first appeared.
- Journey replay steps through what the test user saw and did, with screenshots and per-step accessibility and speed checks.
- Launch checks show SEO, security, accessibility and performance, kept apart from what the test user observed.
- Fix these first lists the fixes in priority order, P01 first.
- All findings are marked high, medium or low, each with a suggested fix and the evidence behind it.
From a report you can share a public link, export findings as CSV, save a PDF, email it to yourself or delete the run.
On the Plus plan, add your name, logo and color under Settings, Branded PDF reports. Save PDF then prints your cover page and color with no Walkthru branding, ready to hand to a client. The report on screen does not change.
If a run was interrupted, for example because the tab closed, it stays "Running". Select End and report to keep its steps and build a partial report.
AI answers: are you named and cited?
On paid plans, AI answers asks AI the questions your buyers ask, then checks in code whether the answer names your brand, cites your pages, and which competitors it names instead. Start with your site and up to five competitors; prompts are suggested from your homepage and you can change every one.
- Groq with web search searches once per prompt. Citations count only when an answer reference matches a returned source. Retrieved pages alone are not citations.
- Gemini, from memory (Plus) answers from what the model already knows. It shows whether AI knows your brand at all, so it counts mentions only.
- Google AI Overviews, Google AI Mode, ChatGPT, Perplexity and Claude are not measured. API samples can differ from consumer apps.
Share of mentions is your share among the tracked brands named in completed answers. Citation rates exclude memory-only, incomplete and older measurements. Expand an answer to inspect its source links and attribution evidence; filter by engine or prompts that name your brand.
Launch Pack checks once, Pro schedules weekly, Plus schedules weekly on both engines. The page shows shared quota and queued work. Answers may take several days; retries use quota too. You get a notification when a set is done. Each sample reports what the AI said on that day; nobody can promise a citation.
Weekly watch and deploy hooks
On the Plus plan, Watch rechecks up to five of your sites every week: SEO, AI search readiness and passive security, the same checks as an Instant Scan. The first check sets a baseline; after that you hear from us only when something changed, for example a deploy that blocked AI crawlers, dropped structured data or reintroduced a missing header, and when a fix landed.
To check right after every deploy, create a deploy hook for the site and send a POST to its URL from a Netlify deploy notification or a step at the end of your CI:
curl -X POST https://YOUR-WALKTHRU-API/hooks/deploy/wh_...A site is checked at most once every 10 minutes, so a burst of deploys costs one check. Journeys are not part of watch; run those from the extension.
Competitor side by side
On paid plans, Compare checks your site and up to three competitors. Switch between score bars and findings by severity, select a metric, inspect measured gaps and open the supporting report. The detailed table retains every measurement and first impression.
No extension or ownership verification is needed. Every site, including yours, gets the same public-only scope. No signed-in browsing, form submissions or owner-only security checks. Page counts describe coverage and are not ranked. Missing measurements are not zero. Each site counts toward your daily scan limit; blocked or opted-out sites remain unavailable.
Connect your editor (MCP)
On the Plus plan, Claude Code, Cursor and other MCP clients can use Walkthru directly: scan a site, read a report, pull the fix prompt into the codebase they are editing, prove the site is yours, open fix pull requests, track AI answers and watch the site after each deploy.
- Open the MCP page and create an API key. Copy it right away: it is shown once.
- Add the server to your editor with the command or file that page shows you, for example in Claude Code:
claude mcp add --transport http walkthru https://api.trywalkthru.com/mcp --header "Authorization: Bearer wt_..."Reports: scan_site, get_report, get_fix_prompt, get_finding, verify_finding, rerun, list_runs, compare_sites, accept_finding, reopen_finding and share_report (the public link and the Launch Ready badge). After fixing one finding, verify_finding re-checks just that finding on its pages, which is faster than a full rerun.
Ownership and GitHub: get_site_verification gives the meta tag your agent adds to the site's <head>, with where it goes in Next.js, Vite, Astro and other frameworks; after you deploy, the same tool confirms it. list_github_repos and open_fix_pull_request preview, then open, a pull request with the config fixes (connect GitHub once in Settings first).
AI answers: track_ai_answers, get_ai_answers, set_ai_prompts and check_ai_answers_now. Weekly watch: watch_site, list_watched_sites, check_watched_site_now and create_deploy_hook (a URL for your CI to call after each deploy). get_plan shows what your plan includes.
Every tool follows the same plan rules, limits and honesty rules as the website, and each key can reach only your own runs and sites. Deleting sites, prompts or connections stays in the dashboard. Journeys still run from the Chrome extension; rerun repeats the server-side checks.
Revoke a key in Settings the moment you stop using it or think it leaked.
Team workspaces
On the Plus plan, Team gives your team, or your client, one shared place: the reports you share, a findings board where each problem gets a status and an owner, comment threads on reports and findings, and a chat that stays with the work. People you invite do not need a plan of their own.
- Roles. The owner runs the workspace on their Plus plan. Admins invite and manage people. Members share reports and triage findings. Viewers read, chat and comment, which suits clients.
- Invite by email for one person: the invitation works only for that address, once they confirm it, for 7 days. Or make an invite link with a code people can type, a number of uses, an end date and, if you like, your company's email domain.
- Share a report with Share to workspace on its page, or turn on auto-share in the workspace settings so every new run, scan and watch check lands there.
- Ask Scout. Write
@Scoutand a question in the chat, for example "what is still open on shop.example.com?". Scout answers in the thread from this workspace's shared reports, findings and recent messages only, and says which report it used. It is an AI answer, so check the report before acting on it. Up to 50 questions a workspace a day. - Findings board. One row per problem and site across all shared reports. A finding marked fixed that shows up again in a later report is flagged "Found again after a fix".
A workspace has 3 seats, counting open email invitations. If the owner's Plus plan ends, the workspace stays readable but pauses chat, sharing, triage and invitations until the owner renews or hands it to a member on Plus. Reports someone shared stay when they leave; deleting a run removes it everywhere.
Your data
- Emails, long numbers and typed values are masked in your browser before anything is sent.
- Screenshots mask form fields and are deleted 30 days after each run.
- Runs and reports are kept until you delete them. Export or delete everything from Settings.
Full details are in the privacy policy.
Troubleshooting
- The side panel says "Not connected"
- Open your dashboard in the same Chrome profile and select Connect extension.
- "Session expired"
- Sign in to the dashboard again and reconnect the extension.
- "Extension not found" on the dashboard
- Check that the extension is installed and enabled in
chrome://extensions, then reload the dashboard. - Screenshots were not captured
- Chrome's screenshot permission was declined. Start a new test and allow it. The run itself still works without screenshots.
- A run is stuck on "Running"
- Open it from the dashboard and select End and report.
- Still stuck?
- Email [email protected] with the report link and we will take a look.