Microsoft Clarity: find friction before an enquiry is sent

A visitor opens the page, reaches the form and disappears. The number of enquiries alone cannot tell you whether the problem was a button, an explanation of the terms or the person's intentions. Microsoft Clarity can help locate observable friction: where people clicked, how far they scrolled and what happened around an attempted submission. An observation, however, neither reveals motivation nor establishes why someone left.

This intermediate guide is for advertisers with their own landing page and authorised access to its analytics. The outcome is one testable hypothesis, a correction task and a rule for judging the result. The wider journey from source to page to conversion is covered in the funnel audit. Here the focus is Clarity screens and the quality of behavioural evidence.

Documentation was checked on October 9, 2026. All numbers and both examples are educational, not results from AffiliateCore or Microsoft customers. No private Clarity project was opened; the tables do not depict its interface. Clarity is a separate service. This article does not claim it is built into AffiliateCore, installed on your site or automatically connected to enquiries and payments.

1. Prepare access and define an enquiry

You need permission to analyse the site, access to the correct Clarity project, someone to verify installation and a form owner who can explain submission confirmation. If there is no project, consult the official getting-started guide. Installing analytics is a separate authorised technical task, not something this article performs. Microsoft also excludes use on sites aimed at people under 18.

Choose one form for an ordinary product, such as a demonstration request for a workplace service. Define a successful enquiry as server-confirmed acceptance, excluding agreed tests and duplicates. Clicking Submit is an attempt, not proof of acceptance. Do not label a visit abandoned if the person completed the task through another intended route, such as an approved contact option.

Record the URL, language, page version, device, period and source of submission confirmation. If form events are not reliably connected to sessions, do not promise an exact abandonment rate from recordings. Clarity can then identify friction while successful submissions are assessed separately through the verified form log. Missing instrumentation is a task of its own.

2. Check privacy before reviewing behaviour

Open Settings → Masking with the responsible administrator. The masking documentation explains that input contents are hidden and setting changes do not repair earlier recordings. First inspect a new authorised test using synthetic data. Never disable masking to read an address, telephone number or enquiry text.

Check more than the form: success messages, URLs, query parameters, headings, expanded cards and custom tags may also expose personal information. Our recommended standard for this task is to keep private data out of analytics altogether. Do not copy enquiry contents into tags, event names, notes or public screenshots. If you find disclosure, stop viewing and report it to the person responsible for data handling.

A banner alone does not establish that consent signals are being passed. Microsoft describes regional requirements in Consent Management; verify applicability and the actual transmission of the user's choice. Masking does not replace consent, and an absence of cookies should not automatically be treated as permission for any observation. Share an anonymised issue summary rather than a private session link with external recipients.

3. Fix one segment in Filters

In the correct project, open Heatmaps and then Filters. Start with the exact visited URL, one page language, Mobile and a bounded period, such as the last complete week. This is an example research window, not a mandatory minimum. Select language through the corresponding page; do not assume browser language equals content language. Add a browser or source if necessary, but avoid splitting a small dataset into isolated observations.

Clarity filters are available in heatmaps and recordings. Re-read the active conditions when moving between screens. Save the working combination as a segment and record its description. “Form, mobile, version 3” is more useful than “bad users”: it describes an observation without judging people.

Keep different form versions and page templates separate. A redesign halfway through the week makes a calendar comparison mix interfaces. Record the change time and restrict the window. Compare the selected page with the actual landing page: the same path with different parameters may show different content. Do not combine those states simply to create a more impressive map.

4. Use maps to ask where, not why

Open the Click map and locate elements around the form. Read names and counts rather than relying on colour. In a click map, select an element and use View recordings to inspect related sessions. One person can click repeatedly, so click counts must not be presented as lost enquiries.

Next open the Scroll map and locate the depth where the form begins. A scroll map describes depth reached, not text read. Reaching a block does not establish that the person noticed a label or understood a condition. Low reach to the form does not automatically mean it should move up: some visitors may already have received their answer.

This is an educational decision table. Heatmap limitations include volume limits and the representation of dynamic content. Check whether the screenshot matches the relevant page state. A popup or nested form cannot be assessed solely from the main-page image.

5. Review recordings using a rule chosen in advance

In Recordings, first check the saved segment. For practice, you might take 20 consecutive available recordings and separately inspect 10 showing a particular signal. These are working quantities for the example, not a statistically sufficient sample. The consecutive list supplies background; the targeted selection helps investigate a mechanism. Do not merge their denominators or call a selection random when it was not randomised.

A Clarity recording reconstructs the page and actions rather than filming the screen. An unusual appearance may require a playback check, not an immediate product change. Watch before and after the event: the last click alone rarely explains what happened. A cursor near a field does not prove attention or intention.

In the observation note, retain only an anonymised reference, page state, action, visible response and alternative explanation. For example: “Button pressed; confirmation not visible; response may be outside the viewport.” Separately record whether the behaviour was reproduced in an authorised test environment. Never transcribe user input or try to identify the person.

6. Separate signals, selection and missing data

Rage clicks and dead clicks are prompts for investigation, not diagnoses. Their meanings are described in the signal documentation. Rapid repeated clicking can accompany difficulty, but you cannot automatically conclude that the person was frustrated or abandoned the form because of it. Look for an observable response and a reproducible sequence.

The Microsoft FAQ explicitly says recordings may not be available for every session. Do not turn a no-sampling claim into a claim that your investigation is complete. Three limitations differ: which visits were observed, which data the tool currently makes available and which recordings the analyst selected. Consent, script blocking and installation faults also require coverage checks.

Do not compare a fresh selection with an older reduced archive as though both were equivalent; consider recording retention. Write “among reviewed recordings” when that is all you inspected. A missing recording does not prove no visit occurred, and no visible submission is not confirmed abandonment. Limited coverage demands caution when estimating prevalence.

7. Educational example: repeated attempts to submit

A fictional service team reviews 20 consecutive available mobile recordings. Six show repeated Submit clicks without a noticeable response. That is 6 / 20 = 30% of the reviewed selection, not 30% of all visitors or lost enquiries. Another 10 signal-selected recordings are reviewed separately; they cannot be added to estimate overall prevalence.

In an authorised test environment with fictitious data, the team finds that a missing-required-field message appears above the current viewport. This is an assumption of the educational example, not an actual discovered defect. The hypothesis is that the person cannot see why the form failed to submit. Alternatives include a slow response, a playback problem or an accidental repeat click.

Choose one change: show a clear error beside the relevant field and move focus to it accessibly. Do not simultaneously change price, field count and advertising. The technical criterion is an error that is visible and keyboard-accessible. The behavioural criterion is fewer repeated attempts without feedback. Server acceptance and the absence of duplicates remain separate checks.

8. Educational example: a form beyond typical scrolling

On another fictional mobile page, the map shows a decline before the form. Of 15 recordings selected for review, nine do not reach that block. The calculation 9 / 15 = 60% describes only reviewed recordings. It is not calculated from map colour and is not a site-wide abandonment rate. Some readers may have obtained what they needed earlier.

Inspect the sequence: a long explanation, a link to conditions and then a demonstration request. The proposed hypothesis is that a reader ready to enquire cannot easily find the next step. Make one change: add a clearly labelled “Go to the form” link after the answer to the main question. Do not remove conditions or interrupt reading with an intrusive popup. People must remain free simply to get information.

Decide in advance whether the proportion starting the form increases among comparable observed visits. Also monitor confirmed submissions, duplicates and irrelevant enquiries. More clicks on the new link without useful enquiries do not establish improvement. If data are sparse, the result is “continue observing”, not a success story.

9. Turn the finding into one task

A correction card needs the observation, hypothesis, alternatives, change, owner and rollback check. Prioritise by reproducibility, severity of the obstacle and coverage of the confirmed segment. A red patch does not make an issue urgent on its own. A reproducible submission failure matters more than a cosmetic guess even when fewer recordings show it.

After preparing requirements for your own landing page, you can register as an AffiliateCore advertiser. Registration does not connect Clarity, automatically approve a product or guarantee enquiries. Partner materials and selection are covered separately in the first-partners guide.

10. Measure the change without false causality

Before changing anything, record the primary metric, denominator, observation window and guardrails. One possibility is server-confirmed unique submissions divided by eligible page visits, but only when that connection has been technically verified. Otherwise keep server enquiries and behavioural notes separate rather than dividing incompatible counts. Duplicates and irrelevant enquiries must not disappear from the assessment to make the result look better.

Causal inference is better supported by a properly planned randomised experiment with comparable groups, adequate volume and a stopping rule set in advance. Clarity supplies behavioural context; this guide does not assume it automatically organises that experiment. In a simple before-and-after comparison, check sources, devices, weekdays, versions and seasonality. A changed metric remains an observed association, not a proven effect.

Do not stop after the first encouraging day. Agree the smallest useful change and duration with the responsible analyst using the baseline volume; there is no universal session count here. Stop a change immediately if it breaks the form or exposes data. An obvious defect need not remain unfixed for an experiment, but sales growth cannot be attributed to its correction without verification.

11. Close the review with a short decision

Before closing, check that the segment is described, the screenshot matches the page, privacy has been checked, observations are separate from explanations, both selections are counted separately, only one change is tested and acceptance is confirmed. Assign an owner. Choose explicitly: keep the change, roll it back or collect more data. “Insufficient data” is a valid result when the missing evidence is specified.

For an existing account, use the advertiser sign-in to apply agreed requirements when working on your own offer. Clarity analysis remains in your separately authorised project; partners do not need private session recordings. A usability report does not replace offer admission or current platform terms.

Keep the date, filters, versions, anonymised observations and final check. Revisit the issue after a page or source change rather than reacting to an old map's colours. Clarity's value here is helping locate a testable improvement. Explaining causes and establishing results require additional work that an impressive session replay cannot replace.