Loading tutorials…
Loading tutorials…
Clarity AI Insights uses GPT-style models to summarize session patterns weekly. Out of the box it is 70% noise. With the right config and a 5-minute weekly triage, it becomes the cheapest CRO analyst you have ever had.
Who this is forOperators with Clarity already collecting recordings + heatmaps who want a weekly digest of 'what changed' without watching 50 recordings. Best for sites with 5,000+ weekly sessions — below that, AI Insights are too sparse to be useful.
What you'll need
Step 1
Settings → AI Insights → Enable. Choose Recordings, Heatmaps, or Both. Pick the pages AI should focus on.
Settings → AI Insights → toggle "Enable AI summaries" to ON.
Scope: "Both" (Recordings + Heatmaps) is the default and best for most. "Recordings only" if you have heavy traffic and want focus. "Heatmaps only" almost never makes sense.
Focus pages: by default AI analyzes all pages. Narrow to your top 5-10 highest-traffic, highest-revenue pages — pricing, checkout, signup, key landing pages. AI works better with concentrated data.
Pages to exclude: blog, docs, /admin, /thanks-you pages (low-CRO-value). Add them in "Exclude pages" — they will still be tracked but AI will not waste cycles on them.
Save.
Step 2
Settings → AI Insights → Weekly digest → recipients, day, summary depth.
Settings → AI Insights → "Weekly digest email."
Recipients: at least one human who will actually read it. Use a real email, not noreply@.
Day of week: Monday is default. Switch to whatever day you start CRO work. Most teams use Monday or Tuesday — gives you the week to act on findings.
Summary depth: "Standard" (3-5 insights per week) or "Detailed" (10-15 insights per week). Start with Standard — Detailed is overwhelming for the first month.
Subject line preview: confirm it includes your Project name so emails do not get filed as "Microsoft updates" and ignored.
Save. First digest arrives 7 days after enable.
Step 3
Each AI Insight has thumbs up/down. Tag at least 10 in the first 2 weeks to tune the model to your priorities.
When the first digest arrives, click into each insight.
For each: thumbs up if actionable + relevant. Thumbs down if noise (irrelevant page, well-known issue, false pattern).
Aim for 10+ ratings in the first 14 days. The model uses your ratings to weight future insights toward what you find valuable.
Common noise to thumbs-down: insights about /admin pages, /api routes, low-traffic pages, or repeated insights about the same minor bug.
Common signal to thumbs-up: rage-click clusters on revenue pages, scroll cliffs on conversion pages, form-abandonment patterns.
After 30 days of ratings, the digest noise level drops noticeably.
Step 4
AI Insights → Funnel analysis. Define a 3-5 step funnel. AI surfaces where users drop and why.
In the AI Insights tab (left sidebar, separate from Recordings), click "Funnels" or "Funnel analysis."
Define your primary funnel: 3-5 steps from entry to conversion (e.g., Homepage → Pricing → Signup form → Submit → Welcome).
For each step, AI computes drop-off rate and clusters the dropping users by behavior pattern.
AI annotates each drop-off step with a hypothesis: "Users on this step rage-clicked the submit button" or "Users on this step scrolled to bottom without engaging."
These hypotheses are AI-generated and 60-70% accurate. Treat as candidates, validate with recordings before acting.
Run one funnel per major conversion path. Do not configure 10 funnels — you will not have time to act on them.
Step 5
5 minutes every digest day: scan insights, pick 1-2 to validate via recordings, archive the rest.
Block 5 minutes on the digest arrival day (Monday morning).
Open the digest email. Read insight headlines only at first — not full descriptions.
Pick the 1-2 that match a known business priority or surprise you. Click in.
For each picked insight: open the linked sessions, watch 1-2 at 4x speed (you already learned this routine in the recordings tutorial).
If the insight checks out → log it in your CRO hypothesis backlog with the date. If it doesn't → thumbs down and move on.
Do NOT try to act on every insight. AI surfaces 5-15 patterns/week. Most weeks, only 1-2 are worth pursuing — that is the point of triage.
Step 6
Monthly: review the AI Insights log, pick top 3 patterns to A/B test, ship tests.
At the end of each month, review your CRO hypothesis backlog (from step 5).
Rank by: (a) frequency in AI Insights (patterns flagged 3+ times beat one-offs), (b) page revenue/conversion value, (c) implementation effort.
Pick the top 3 hypotheses. Convert each into an A/B test brief: hypothesis, variant design, success metric, sample size needed.
Ship at most 1-2 A/B tests at once to avoid traffic-splitting issues on lower-traffic sites.
After test completion, log the result in the backlog — wins teach the team what to test next; losses teach what NOT to test.
Common mistakes
Enabling AI Insights on a site under 5K weekly sessions
What goes wrong: AI clustering needs sample size. Below 5K sessions/week, insights are random noise — 'users on /about scrolled less than average' style patterns that mean nothing. You waste 30 min/week reading meaningless emails and conclude AI is useless.
How to avoid: Below 5K weekly sessions: disable the digest. Use heatmaps + recordings + manual analysis instead. Re-enable AI when traffic grows.
Trying to act on every insight in the digest
What goes wrong: 10-15 insights/week × 30 min each to investigate = 5-7 hours/week. You stop reading the digest. The investment in AI Insights becomes pure cost with zero return.
How to avoid: 5-min triage: scan headlines, pick 1-2 to validate, archive the rest. The signal is in the patterns that repeat across weeks, not every individual flag.
Not training the model with thumbs up/down
What goes wrong: Default AI Insights are tuned for 'general Clarity users.' Without your training data, the digest stays generic and noisy forever. The 30% signal:70% noise ratio never improves.
How to avoid: Tag 10+ insights in the first 2 weeks. After 30 days of tagging, noise drops noticeably. Treat the tagging as the cost of having the feature.
Confusing AI hypotheses with conclusions
What goes wrong: AI says 'users on /pricing are rage-clicking the submit button.' You assume this is true and ship a fix → A/B test flatlines → you waste $1,500-3,000 on the redesign + 2 weeks.
How to avoid: Always validate AI hypotheses by watching actual recordings BEFORE building the fix. AI is a hypothesis generator; recordings are the validator; A/B tests are the proof.
Configuring AI on every page (no exclusions)
What goes wrong: AI analyzes /admin, /api, /docs, /blog — wasting cycles on pages that have nothing to do with conversion. Digest is full of irrelevant insights about your help center while ignoring your checkout funnel.
How to avoid: In Settings → AI Insights, explicitly include only your conversion pages (top 5-10) and explicitly exclude support/admin/blog. Narrow scope = higher signal density.
Recap
Done — what's next
How to set up Clarity session recordings without drowning in noise
Read the next tutorial
Hand it off
AI Insights are most useful in the hands of someone who knows what to do with each pattern. A CRO specialist treats the weekly digest as a starting backlog — validating 1-2 patterns/week and shipping 1-2 A/B tests/month. EverestX matches you with a vetted specialist starting at $14-16/hr.
See CRO specialist rates
Microsoft's public privacy policy says aggregated, anonymized behavior patterns may be used for model improvement. Your raw recordings and PII are not used. If you have strict data governance requirements, disable AI Insights — the rest of Clarity (heatmaps, recordings) does not invoke AI.
GA4 Intelligence flags traffic and revenue anomalies (metrics). Clarity AI flags behavior anomalies (patterns in clicks, scrolls, rage events). They are complementary — GA4 tells you what changed in conversion rates; Clarity tells you what changed in user behavior.
Yes — the digest email can have multiple recipients (Settings → AI Insights → Weekly digest). Individual insights link back to Clarity where Team members with Viewer+ access can drill in. Avoid forwarding insights as screenshots — context is lost.
Quarterly. Your top conversion pages change as you launch new products or campaigns. Set a calendar reminder to revisit Settings → AI Insights → page scope every 3 months.
Thumbs down the insight as 'resolved.' AI uses 14-30 days of trailing data, so even after a fix ships, old data may still trigger the pattern for 2-4 weeks. Be patient — the flag fades.
Microsoft Clarity
Session recordings are powerful — until you have 50,000 of them and no time to watch any. The difference between useful and useless is filter discipline. Here is the system that surfaces the 10-20 sessions per week that actually inform decisions.
Microsoft Clarity
Rage clicks and dead clicks are the most actionable signals Clarity gives you — every one represents a user actively frustrated with your site. This walkthrough shows the diagnostic sequence to fix them.
Microsoft Clarity
GA4 tells you conversion rate dropped 18% last Tuesday. Clarity tells you why. The integration lets you click from a GA4 anomaly straight to the recordings that caused it — but only if the connection is set up correctly.
Microsoft Clarity
You installed Clarity. You watched some recordings. You ran an A/B test or two. So why is conversion rate still flat? This is the honest framework: when DIY CRO hits its ceiling and a specialist starts paying for themselves.