Beta. Signals is in Beta. It works today and the ideas are real,
but the interface and wording on these pages can still change as we
refine it. If something reads differently in your dashboard, trust the
dashboard and email support@abtestly.com so we can fix the page.
The four steps
Scan a page
A browser extension captures the page from your own live site:
desktop, mobile, and mobile again on a throttled connection. Walk
several steps of a funnel in one pass.
Connect Google Analytics
One click, read only. Lets a scan read your funnel data and credit
it on the ideas it speaks to.
Read a scan
The idea board. Ranked by a PXL score, grouped into campaigns,
triaged New, Approve, or Dismiss.
Build a test from an idea
Build this creates a real experiment draft, prefilled from the
hypothesis. Its result feeds back into the next scan.
The evidence, and honesty about it
This is the part that matters most, so read it before you read a scan. Every scan reads the page structure. That is the DOM: the headings, buttons, forms, copy, and layout the browser actually rendered. Structure is the first signal, and it is always present. It is enough to spot a buried call to action, a form asking for too much, a value proposition that never appears above the fold. Connecting Google Analytics 4 lets a scan read a second, independent signal and credit it on the ideas it speaks to: behavior. What visitors actually do. Where they fall out of a funnel, how conversion differs by device, where revenue comes from by channel. Structure tells you what the page is. Behavior tells you what is happening on it. Customer qualitative voice is the other route to a second signal. What your customers said in their own words, supplied by you in Brand context: reviews, support themes, survey verbatims. An idea it is credited on can reach Test ready with no GA4 at all. The first party heatmap is a fourth source, but it corroborates funnel data rather than standing in for it, so it counts only alongside a credited GA4 answer. The evidence contract sets all four out. An idea is marked Test ready only when a second independent signal confirms what the structure suggests and its score clears the threshold for its evidence tier. Until then it reads Needs evidence, Below threshold or Unverified, as the badge table sets out: an honest finding for you to check, not a promise that it will win. Signals never invents a statistic when the data to support it is not there. When the number exists, you see it. When it does not, the idea says so plainly.The closed loop
Most audit tools stop at the report. They never see whether their idea worked, so they cannot get smarter about your site. ABTestly owns both ends, the idea and the experiment, so it can close that loop.Evidence in · three parallel streamsReasoning · narrowed to a ranked backlog
into 01 · Intakefrom 07 · OutputOutcomes feed back as priors for the next scan
A few minutes · end to endIt produces the backlog · it never runs your test
Inside a scan, step by step
Inside a single scan, the audit works through seven stages, from the URL you hand over to the ranked backlog it hands back. Here is what happens at each one.01 · Intake
You start by giving ABTestly the URL you want audited, plus any context you have: your niche, average order value, the goal that matters, your device split. How you connect your data decides how far the scan can go. Connect GA4 and a scan can read real behaviour. With neither GA4 nor customer voice in Brand context, every idea stays in the Structural tier, working from the page structure alone. What reaches a scan sets what it can score, not what any one idea earns: a connected source that returns nothing reaches the scan with nothing to score, and each idea’s own evidence tier is settled later, from the evidence that actually bears on it. Intake also confirms the page is on your own site before it runs.02 · Capture
A browser extension captures the page the way a real visitor sees it, from your own logged in session, so the scan reads the rendered page and not a bare URL. It picks up more than twenty structural signals, from page type and heading order to form fields, calls to action, review counts and trust badges. Three screenshots follow, on desktop, on mobile, and on a throttled connection, each reconciled against what it read. It only ever reads: it never submits payment and never tries to get around a site’s bot protection.03 · Behaviour
Where GA4 funnel data reaches the scan, it can read where real visitors actually drop off. That is the behavioural half of the two signal rule: funnel drop off, conversion by device, revenue by channel, plus new against returning visitors and your busiest pages. It matters because a finding is only promoted once a second independent signal agrees with what the page showed, the two signal rule. GA4 funnel data is one such signal; the customer voice you supply in Brand context is the other.04 · Payload
Everything gathered so far is assembled into a single prompt the model can reason over, so it works from real evidence rather than a bare URL. That prompt carries the three screenshots next to a block of structured text: your business context, the page signals, the live picture from GA4, and any earlier test results from this same site. It is framed as evidence to weigh, not a list of claims to accept.05 · Generation
Now the model proposes ideas that speak to the problem you named, each grounded in a behavioural mechanism and evaluated against the evidence, not guessed. Every idea arrives as a structured hypothesis with 18 fields behind it, among them the effort, a research citation, a tracking spec, and its answers to the eight evidence questions ABTestly scores. The model answers those questions; it never sets the score, the tier or the rank. Each idea sits under one of five CRO pillars and is read through a named model such as Fogg or Cialdini. When the evidence is thin the idea is flagged rather than dropped, with the single cheapest check that would confirm it, the two signal gate.06 · Falsification
Before an idea reaches you, it goes through a pass that argues against it. This adversarial step names the most likely way the idea could fail, the confounder most likely to mislead you, another explanation for the same evidence, and the one test that would tell them apart. You get that counter argument attached to the recommendation. The pass never blocks a scan: when it cannot complete, the ideas still reach you, without that counter argument attached.07 · Output
What comes back is a ranked backlog. ABTestly scores every idea from the model’s answers and orders the list by that PXL score, so the best return for the effort sits at the top, and each score is read against the maximum that scan could reach rather than a fixed scale. Each item carries its evidence tier, a rough size for the opportunity, the effort, a suggested run time, and the tracking you would need to set up. There are four evidence tiers, from verified down to purely structural, and the tier belongs to the idea rather than to the scan, so two ideas from one scan can carry different tiers and you can see how much weight each recommendation should carry. Signals writes the backlog and the plan; it never starts the test for you, so what ships stays your call.What Signals does not do
- It does not run tests on its own. Nothing goes live until you click Build this, fill in the details, and start the experiment yourself.
- It does not fabricate numbers. No statistic appears unless the connected data supports it.
- It does not need your passwords. The scan runs in your own browser, on your own logged in session. See Scan a page.
Where to go next
Run your first scan
Install the extension and capture a page in a couple of minutes.
Connect Google Analytics
One click, read only. Lets a scan read your funnel data and credit
it on the ideas it speaks to.