Clicks but no conversions: a practical funnel audit

People follow the link, yet very few complete a useful action. It is tempting to change the audience, make the button louder or buy more traffic. The same outcome can have very different causes, however: the page never became usable, the offer changed after the click, the conditions were unclear or the form could not be submitted. Mixing these cases makes every change a guess.

This guide examines the journey before an action happens. If someone has already completed the action but it is missing from a report, use the separate missing-conversion checklist. Postbacks, attribution disputes and payment holds are outside this audit. Our fictional examples concern a team planning template and bicycle repair bookings. They are general educational examples, not live AffiliateCore offers or promises of improved performance.

Your output is a short list of evidenced obstacles and one testable next step. A simple worksheet is enough. You do not need to buy a new tool or record every visitor to begin understanding the journey.

1. Define the action that actually matters

Write one sentence: “The visitor voluntarily requests a demonstration of the planning template, knowing that the response will arrive by email.” This identifies the action, its subject and what happens next. “Increase engagement” is too vague to audit. If colleagues disagree about the outcome, settle that definition before comparing numbers.

Separate opening the form, submitting it successfully and producing a usable enquiry. For a repair shop, a usable enquiry might concern a supported bicycle type within its service area. Spam, duplicates and empty requests do not become useful merely because a server accepted them. Write the criteria before the review, not after disappointing results.

Specify the observation period, time zone and time needed to assess the outcome. Yesterday's enquiries may still be awaiting review; last week's may already be fully assessed. They are not equivalent groups. Mark missing information as “not measured”. Zero means the event was observed and did not occur, which is a different statement.

Assign responsibility at each stage. A publisher can change their own copy but may not control the destination form. For a page owned by someone else, prepare reproduction steps and a question for its owner. Lack of access is not evidence that the page works correctly.

If you plan to work through AffiliateCore, you can register as a publisher and prepare your profile. Then, if a suitable permitted non-gambling offer is available, compare its requirements with the action defined above. Already have a profile? Sign in. Registration does not guarantee offer access, approval or results. If nothing suitable is available, continue auditing your own page without launching anything.

2. Map the journey and name every denominator

Educational diagram 1: message and promise → usable landing page → understood offer → form started → successful submission → suitability review. The last stage protects quality, but it should not conceal friction earlier in the journey.

List the observations actually available. An advertising click counter and a site visit counter can measure different things: repeated clicks, sessions or unique visitors. Their difference is not automatically the number of people lost because of page speed. First establish what each counter includes and excludes.

All figures below are invented. The example assumes that an educational measurement system connects stages for the same group, removes repeats under an agreed rule and uses a consistent observation period. Confirm those conditions before calculating a comparable funnel in a real audit.

The suitable-enquiry rate is 12 / 400 = 3%. Submissions divided by landing page openings are 20 / 320 = 6.25%. Neither number is an industry benchmark. A large gap around the form suggests a place to investigate, not a proven cause. The table cannot tell you that a different button would generate another 60 enquiries.

Where stages cannot be connected, keep the counters separate and perform a manual walkthrough. Do not invent missing stages. A successful page response alone does not establish that a person saw or understood its main content.

3. Compare the promise with the first screen

Place the original message beside the landing page. Write down five elements: the intended audience, the problem, the promised result, material conditions and what the button will do. Find clear support on the page for each one.

Consider an advertisement offering a downloadable planning template. The destination only offers a phone call with sales staff. Even a beautiful, fast page breaks the expectation. An honest correction is to provide the promised file or change the original message to offer a consultation. Adding “free” to the button does not resolve the mismatch.

Check limitations too. If the template requires a particular editor, say so before the form. If the demonstration takes place at an agreed time, do not call it instant access. If the next step involves payment, do not hide the price and essential conditions behind a vague “continue”.

Ask someone unfamiliar with the project to read the first screen and explain what they expect next. Avoid coaching them towards your intended answer. Note where their interpretation differs. This is a qualitative clarity check, not a representative demand study or a conversion forecast.

4. Walk through the mobile route from the beginning

Open the actual address used in the placement, including its redirects, rather than just a bookmarked destination. Use your own device and an authorised testing environment. If submission creates a real operational enquiry, agree on a test mode and exclusion from operational counts first.

Record the device, browser, screen orientation and connection type. Start without saved authentication, then repeat a typical returning visit. Check that the language remains appropriate, the headline matches the message and information panels can be dismissed normally.

Read at the default text size and again with enlarged text. Open the keyboard in every field. A lower button must remain reachable rather than disappearing behind the keyboard or a fixed panel. Go back one step and check whether entered information survives and whether any required repetition is explained.

A slower connection can help reproduce waiting, but it does not replace evidence about your actual audience. Repeat the failing step and record the sequence, not just the final screenshot. If a failure occurs once in several attempts, report that honestly. One device fault does not make an entire mobile audience low quality.

For a technical starting point, consult Google's Web Vitals overview, which covers loading, responsiveness and visual stability. Good technical results do not establish that visitors understand or want the offer.

5. Treat the form as a task in its own right

List every field and explain why it is needed at this moment. Sending a file by email does not itself explain a request for a telephone number. If the number serves another necessary process, disclose that purpose clearly. Remove genuinely unnecessary fields while retaining information needed to fulfil the request.

Review field labels separately from text inside empty fields. W3C's guidance on labels helps preserve the meaning of a control during entry. Ask a developer to check accessible names where a visual inspection is insufficient.

In the authorised test environment, submit an empty form, introduce an email error, correct it and resubmit. Is it clear which field needs attention? Does the other input survive? Is success clearly acknowledged? W3C's notification guidance addresses feedback for errors and successful actions.

The button should describe what really happens. “Request a repair appointment” is more accurate than “Confirm my appointment” when staff still need to agree a time. After submission, explain the next step and the actual response window. Agree that window with the team handling enquiries rather than inventing an attractive promise.

Also check consent and the ability to decline without pressure. The purpose is an informed action, not an accidental submission. Preselected optional consent, artificial countdowns and hidden obligations are not sensible ways to improve the figures.

6. Separate audience intent from a broken experience

Compare a small number of meaningful groups: mobile versus desktop, page language or the original message. Look first for a substantial, reproducible difference. Dozens of intersections in a small dataset make it easy to create a persuasive story for almost any result.

In the bicycle example, the message promises instructions for adjusting brakes at home, while the destination immediately asks for a repair booking. Readers may want the information but not the service. Separate the instructions from the booking honestly instead of dismissing these people as poor traffic. If service is limited to one area, disclose that before the enquiry.

Watch suitability as well as volume. Removing the service-area explanation might produce more submissions but also more requests from places the shop cannot serve. That increase is not necessarily an improvement. Keep useful restrictions and make them easier for visitors to understand and act on.

With few observations, resist saying that a source “does not work”. State what is missing: enough elapsed time, comparable records or reproduction of the fault. Do not automatically extend spending merely to produce a more impressive report.

7. Turn observations into repair cards

Educational diagram 2: observation → possible explanation → verification method → one change → decision criterion. This sequence stops an assumption quietly turning into an established fact.

A useful card might read: “After email entry on the specified phone, a fixed panel covers the submit button. Reproduced three times. Hypothesis: some visitors cannot finish. Owner: landing-page team. Change: remove the overlap. Verification: repeat the route and check enlarged text for new overlaps.” This is more actionable than “improve mobile conversion”.

Add the date, address, copy version, anonymised evidence and status. Keep email addresses, phone numbers, passwords and visitor messages out of shared audit documents. Screenshots with invented test data will often explain an interface fault adequately. Restrict access to logs when they are needed.

Prioritise confirmed inability to act, then material promise mismatches, then unclear explanations and visual details. Priority depends on reach and consequences, not simply on how easy a button colour is to change. Where a fault is not reproduced, assign a verification task instead of declaring its cause resolved.

8. Check the fix without overstating causation

Before changing anything, record the source, period, device mix, outcome definition and page version. Repeat the same manual scenario after the fix. If the fault disappears, you can report that this specific defect was removed. You cannot automatically credit it with every business change the following week.

For a comparison of variants, choose the main measure, safeguards and observation period in advance. Avoid changing the promise, audience, price and form together. A properly organised experiment needs an allocation that supports comparison. Two different weeks offer weaker evidence because external conditions can change.

Observe suitable enquiries, errors, complaints and handling workload alongside submissions. Do not stop at the first attractive result or claim statistical confidence without an appropriate calculation. “The obstacle is gone; its effect on enquiry rate remains unknown” is a useful outcome when data are limited.

Prepare a fallback for someone who cannot finish the form. You might explain where to report a technical problem nearby. That route should not require passwords, documents or the whole enquiry to be posted in a public message. Name only a channel that is actually staffed; do not promise round-the-clock chat if nobody reads it overnight. Test the fallback as carefully as the main route.

Also record when to restore the previous page version. Removing an overlap is not a successful fix if the submission acknowledgement now fails. The owner should know which version to restore and which colleagues to inform. Rolling back the interface does not mean deleting enquiries already received. Make sure repeated checks do not create unnecessary operational records or send unwanted emails.

9. Finish with a one-page decision

Your final document should answer six questions: what action was checked, which group was observed, where the obstacle appeared, what supports the finding, who owns the fix and when it will be checked again. List unknown information and stopping conditions separately. Keep a full redesign wish list out of this decision.

Practical assignment: choose one ordinary, non-gambling landing page you control. Write its promise and expected action, walk through it on a phone, map the observable stages and prepare no more than three repair cards. Give each card one immediate next step. Remove personal information before sharing it and check that someone else can follow the reproduction instructions.

For this assignment, existing users can sign in to AffiliateCore and, if a suitable permitted non-gambling offer is available, compare its conditions with their page’s promise. Do not invent missing terms or generate operational events for the exercise. If no suitable offer is available, use only the fictional educational example.

For questions about a specific route, contact AffiliateCore with the page address, language, device, time and an anonymised description of the step. This is a request for clarification, not an offer of automatic auditing or guaranteed results. You can also explore the practical article archive to continue your own review.

Ready to use your findings in your preparation? Create an AffiliateCore publisher profile so that, if a suitable permitted non-gambling offer becomes available, you can compare its conditions with your reviewed page. Existing users can sign in. Availability, approval and results are not guaranteed; without suitable conditions, there is no need to launch.

Sources checked on 9 October 2026: the web.dev and W3C documentation linked above. The stage map, figures, repair cards and audit sequence are original educational editorial material. They are not client results, industry standards or descriptions of built-in AffiliateCore features.