AI diagnosis: what to fix first on your site
Once your site has a few days of sessions, the Diagnosis tab reads the behavior of your visitors and writes a short report: what is going wrong, how many people it affects and what to change, ranked by priority. You tell it your goal first, so the report focuses on what matters to you.
How it works
- Fill in the brief: the problem you want to solve, your goal, a bit of context, your main pages and key buttons.
- Pick a period and device and run the diagnosis.
- Read the report: findings ranked by priority, with the evidence behind each one.
What goes into the report
- Funnel drop-offs and the sessions behind them.
- Clicks, rage clicks and scroll patterns by screen.
- Where visits end and how long they last.
- Your brief, so the advice matches your goal.
See Visiflow in action
Short demo videos on sample data (not a real customer). Each one shows a task from start to finish.
Find where visitors drop off
Open the funnel, spot the worst step and see which pages people were on.
- Open Funnel
- Hover the biggest drop
- Click it to see where they went
Videos use fictional sample data.
Private by design
The AI does not receive video and does not receive user IDs. Everything that goes to it is scrubbed first: emails, phone numbers, CPF and long numbers are removed. Note that the text of buttons is recorded as is, so if a button can show personal data, hide it with class="rastro-mask".
Limits you should know
- There is a daily limit of diagnoses per site, and one runs at a time.
- A diagnosis takes a short time: it runs in the background and you can come back to it.
- It is a starting point. Confirm the findings with the heatmap and the replays.
Use it from your own tools
Visiflow has an external API for diagnoses: create one with a key that has the right scope, read the results, or receive them at a callback address signed with a secret. See the API docs and the OpenAPI file.
An example of what a finding looks like
A finding names a problem, says how many visits it affects, shows the evidence and proposes a change. For example: "On mobile, many visits stop between pricing and the trial form. The heatmap of the pricing page shows clicks on the plan names, which are not links, and the replays show people tapping them several times. Proposed change: make the plan cards clickable and move the trial button above the first plan." You then check the funnel step, the heatmap and two replays before you change anything.
How to act on a report
- Read the top three findings only.
- Open the evidence for each one: funnel step, heatmap and replays.
- Reject findings you cannot confirm.
- Pick one change, ship it and measure the same step again.
- Run another diagnosis after a few days of new data.
Who it helps most
- Owners without an analyst, who need a short list of what to fix.
- Agencies, who can start an audit with a report and confirm it in the data.
- Product teams, who want a fresh pair of eyes on behavior every week.
Good briefs make good reports
- Name one problem, not five ("people leave the pricing page").
- State a goal you can measure ("more trials started").
- List your key pages and buttons by name.
- Add context the data cannot show (a campaign, a redesign).
Questions
Which AI does the diagnosis use?
It uses Google's Gemini model, with the data scrubbed first.
Does the AI see my visitors' emails?
No. Emails, phone numbers and long numbers are removed before anything is sent, and user IDs are not sent.
How many diagnoses can I run?
There is a daily limit per site, shown in the dashboard.
Can I trust the report?
Treat it as a prioritized list of hypotheses. Check the evidence in the funnel, heatmap and replays.