What an expert review is good for
An expert review sits early in conversion research. It needs no traffic, no tools and no participants, so it can run on a new journey before launch, on a low-traffic site where recordings would show little, or at the start of a research sprint to tell the other methods where to look. It finds problems that break a written rule: a headline that does not repeat the advert's promise, a form field nobody needs, proof far from the claim it supports, a delivery cost that appears for the first time at payment.
It is weak at everything that depends on how real visitors think. Reviewers know the product names, the pricing logic and the design patterns, so they cannot be confused the way a first-time visitor is, and they tend to flag what breaks their own preferences. The conversion research methods guide treats a finding from a single method as a lead to be checked by another, and a review is a single method. Its output is a list of leads with the evidence each needs, which analytics, recordings, polls and usability sessions then confirm or reject.
Three neighbouring tools do related jobs. The landing page audit scores one page against fixed criteria into bands; the landing page checklist is a pass before a new page goes live; and walking the customer journey yourself has a marketer go through the whole journey as a customer, across channels and beside two rivals. The review covers every step of the on-site conversion path, rates what it finds and stops short of scoring.
The criteria come from the site's own guides, not from a published heuristic set, and each is a question a reviewer can answer by looking at the page.
| Criterion area | What the reviewer checks | Guide it comes from |
|---|---|---|
| Message match | The first screen repeats the promise, offer and wording of the advert, email or link that brought the visitor | Message match |
| Structure and offer | The sections answer the visitor's questions in order, and the offer says what they get, on what terms and what happens next | Anatomy of a landing page |
| Proof and trust | Proof is specific, attributed and placed beside the claim it supports; guarantees and policies are where doubt arises | Trust and proof |
| Forms | Only the fields the next step needs, visible labels, errors in text beside the field, input kept after an error | Forms |
| Speed and mobile | Actions within reach of a thumb, nothing that depends on hover, the page responds promptly to a tap | Speed and mobile |
| Checkout | Guest checkout, delivery cost and date shown before the last step, the usual payment methods, clear recovery from errors | Checkout optimisation |
| Accessibility | The journey can be completed with a keyboard alone and with a screen reader; colour is never the only signal | Accessibility as conversion work |
Before you start
- The anatomy guide, read. The prerequisite is the anatomy of a landing page guide, which sets out what each section of a page is for.
- Two reviewers. People who know the guides but did not design the journey; one reviewer's list is one set of preferences.
- The entry points. The adverts, emails, search results and links that bring most visitors to the journey, with screenshots, so each review starts where visitors start.
- Real devices. The phones and browsers the analytics show visitors using, and a desktop for comparison.
- An issue log. A shared table with columns for location, criterion, issue, severity, confidence, evidence that would confirm it, and decision.
- Test data. Test accounts and payment details, so the reviewers can reach the confirmation without placing real orders.
The steps
Define the journey, the visitor and the entry points
Write down the journey from entry to conversion, such as "advert for a free trial lesson, landing page, timetable, sign-up form, confirmation", and the visitor it is for: who they are, what they already know and what they want. List the main entry points with their promises, because the first screen is judged against what brought the visitor. A review without a defined visitor turns into a review of the reviewers' tastes.
Assemble criteria from the guides
Build the criteria list from the table above and the guides behind it, and add the journey's own rules, such as legal wording the business must show. Write each criterion as a question with a visible answer: "Does the headline repeat the advert's offer?" rather than "Is the page persuasive?" For accessibility, follow the landing page audit's use of the Web Content Accessibility Guidelines, listed in this page's sources: a guide to what to check, organised around content that is perceivable, operable, understandable and robust, as the accessibility guide explains, and not a legal opinion. For hierarchy, proof and form friction, the audit treats the usability research collected in the Nielsen Norman Group's articles, also listed in the sources, as a sound reference point.
Review independently on real devices and entry points
Each reviewer works alone, without seeing the other's notes, and starts from the real entry points on the devices visitors use: the advert on a phone, the email in a mail app, the search result in a private window. Independence is the point; two people who review together converge on one view and lose the second opinion. Each reviewer completes the journey to the confirmation with the test data, then goes back through it step by step.
Walk each step asking what the visitor needs
At every step, ask three questions from the visitor's side: what do they need to know to continue, what do they need to do, and what do they need to trust? Check the answer against the criteria. On a landing page, the visitor needs to know they are in the right place, which is message matchHow closely a landing page repeats the promise and wording of the ad or link that sent the visitor.; to see one clear call to action (CTAThe prompt that tells a reader the one action to take next, such as a button or link.); and to trust the claim, which is proof beside it. On a form, they need to know why each field is asked and to recover from an error without retyping. Note where the page answers a question the visitor has not yet asked, or never answers one they have.
Log issues with location and criterion
Write every issue as a row: the step and the element, the criterion it breaks, what the reviewer saw, and a screenshot. Describe the issue, not the fix: "The delivery cost first appears on the payment step" rather than "Show delivery cost on the basket page". Keep to one issue per row, so that each can be confirmed, fixed or rejected on its own. An issue that breaks no criterion is noted separately as an observation, because it is a preference until someone can say which rule it breaks.
Rate severity and confidence
Give each issue two ratings. Severity uses a scale the team agrees before rating, such as blocks, delays and irritates: whether the issue stops the visitor, slows them, or only annoys them. Confidence is the reviewer's honest view of how sure they are that real visitors meet the issue: high where the page plainly breaks a rule on every device, low where it depends on how a visitor reads a phrase. The pair matters more than either rating; a severe issue rated low for confidence is a priority for verification, not for a fix.
Separate bugs from opinions
Some issues are facts: a price on the page that differs from the advert, a link that leads nowhere, a form that will not submit on one browser, a field that rejects valid input, a control a keyboard cannot reach. These are bugs, and the conversion research hub is plain that bugs are fixed and moved on from, not tested. Mark them and send them to their owners at once. Everything else, such as whether a headline is clear or proof is persuasive, is an opinion until other evidence agrees.
Merge the reviewers' lists
Bring the two logs together and match issues that describe the same problem. An issue both reviewers found independently earns more confidence than one found by a single reviewer; an issue only one found is kept, with its source noted. Where the reviewers disagree about severity, record both ratings and the reasons rather than averaging them. Rank the merged list with the impact-effort matrix, plotting each issue by the effect a fix could have and the work it needs, as a first sort rather than a final ranking.
Name the evidence that would confirm each issue
For every issue that is not a bug, write what would confirm or reject it and which method would show it. A suspected leak at a step is confirmed by the step's rate in the funnel (FunnelModel of the customer journey stages.). Suspected confusion with an element is confirmed by repeated taps in recordings. A suspected missing answer is confirmed by poll answers at that step. A suspected misreading of a label is confirmed by usability sessions. Writing the evidence down makes the review honest about what it does not know.
Feed the backlog and plan verification
Add each issue to the research backlog in the research hub with its location, the review as its source, the evidence still needed and its status. In the research backlogThe single list of conversion research findings, each with its problem, location, evidence, reach, likely cost, status and decision: fix, hypothesise or park., a review issue stays a lead until another method confirms it. Plan the verification: which issues go into the next funnel check, recording review, poll or round of usability sessions, and when. Only issues that another method confirms move on to the hypothesis tutorial and become a hypothesisA testable statement of the change to make, the effect expected on a named metric and the evidence-based reason for expecting it. for testing. Score no page and promise no number of findings; the review's value is the verified list it leads to.
A language school reviews its trial-lesson journey (illustrative)
Lingotuft Language Classes reviews its free trial-lesson journey
Lingotuft Language Classes is an invented online language school that offers a free trial lesson before people buy a course. Its marketing manager asks two marketers who did not build the journey to review it on phones, starting from the school's current search advert, which promises "Your first lesson free, then courses from £12 a lesson", and from the email that goes to newsletter subscribers. Working separately, each walks the journey from the advert through the landing page, the timetable, the sign-up form and the confirmation page, asking what a working adult who wants an evening lesson needs to know, do and trust. Their merged log holds issues at every step, and the three that matter most were found by both reviewers independently. The landing page quotes courses from £15 a lesson, which differs from the advert: a bug, and the page is corrected the same day. The timetable's timezone picker defaults to the school's own timezone and hides evening slots for visitors elsewhere, rated "blocks" for those visitors with medium confidence, because the reviewers cannot tell how many visitors are in other timezones. The confirmation page says "You're booked!" with no joining instructions, rated "delays" with high confidence. The other issues, mostly about the order of the landing page's sections, are opinions rated low for confidence. The team sends the timezone issue to the funnel analysis, splitting the timetable step by visitor country, and the confirmation issue to a short poll on the confirmation page. Neither becomes a test until that evidence is in.
Common mistakes
- Reviewing together. Two people in one room produce one opinion; review independently, then merge.
- Starting from the homepage. Visitors start from adverts, emails and search results; so should the review.
- Reviewing on a desktop only. Problems that live on phones go unseen.
- Criteria nobody can check. "Is it persuasive?" is a preference; "does the headline repeat the offer?" is a criterion.
- Logging fixes instead of issues. A proposed fix hides the problem it was meant to solve.
- Treating opinions as findings. A review produces leads; analytics, recordings, polls and sessions decide.
- Testing bugs. A wrong price or a broken link is fixed at once, not tested.
- Ignoring confidence. A severe issue the reviewer is unsure of needs evidence before it needs a fix.
- Scoring the journey. Scoring bands belong to the landing page audit; the review's output is a verified list.
