Write the question before the change

An SEO test record should help another person understand what changed, why it changed and what the team can reasonably conclude afterward. It does not need to be a long report. It does need to exist before the edit, while the original reasoning is still available. Otherwise, it is easy to turn a later movement in a chart into a story that the team never intended to test.

Start with one decision. For example: should the category introduction explain the product selection more clearly before visitors reach the product grid? That question is specific enough to guide a change. A broad goal such as improve SEO is harder to evaluate because almost any later observation could be used to support it. The record should make the intended decision visible to a colleague who was not in the planning meeting.

The example below is hypothetical. It uses a small retail website and a proposed category-page edit to show how a record can be structured. The same approach can be adapted to other page types, but the evidence and comparison limits will differ. Treat the template as a prompt for careful documentation, not as a formula that makes a website experiment scientifically conclusive.

Describe the present state

Record the exact URLs included in the change. Save the current copy or a screenshot, and note the date and time of capture. If a page template generates the relevant content, identify the template as well as the sample URLs. A later reviewer needs to know whether the edit affected twelve named pages, every page using a template or a conditional subset that depends on product data.

Write down the reason for choosing these pages. Perhaps a support question suggests that customers do not understand the difference between two product groups. Perhaps a content review found that the introduction omits information needed to choose a subcategory. Distinguish those observations from assumptions. A phrase such as customers are confused needs supporting evidence; a phrase such as the team suspects this explanation is unclear accurately labels a hypothesis.

Include practical conditions that could change during the review period. Product availability, promotions and unrelated releases can alter the context. You do not have to predict every event. You do need a place to record events when they occur. A short context field is easier to maintain than reconstructing several weeks of activity after the fact.

Make the hypothesis readable

A useful hypothesis has a change, an expected observation and a reason connecting them. In this example: replacing a generic category introduction with a concise explanation of the available product types may help relevant visitors choose a subcategory; the team will review the agreed page-level observations and customer behavior after release. The exact observation depends on the tools and data the team actually has.

Avoid stuffing several unrelated changes into that sentence. If the team rewrites copy, changes navigation and replaces product photography at once, the record should say it is evaluating the combined release. It should not later attribute an outcome to the copy alone. Sometimes a bundled release is the practical choice. Documentation is valuable precisely because it makes that limitation clear.

State what the test will not resolve. A small set of pages may not represent the entire website. A seasonal category may behave differently from a stable one. The proposed edit may improve clarity even if the available search data is too limited to show a useful pattern. These limits belong in the plan so the final discussion can return to them.

Choose observations you can reproduce

Record the data source, date range and filters before collecting the baseline. Search Console exposes clicks and impressions with breakdowns including query, page and country. If those are the measures being used, save the exact view or export settings so another person can repeat the collection. A label such as organic performance is too vague for a reliable review.

Keep measurements separate from interpretations. A measurement might be the recorded clicks for a named set of URLs over a stated period. An interpretation might be that the new introduction attracted a more relevant audience. The second statement needs more support than the first. Writing them in separate fields discourages a team from treating an attractive explanation as an observed fact.

If there is a comparison group, explain why it was chosen and where it differs. Similar-looking categories can still vary in demand, inventory and promotion. A comparison may provide useful context without acting as a clean control. Label it accordingly. It is better to preserve an imperfect but explicit comparison than to imply that the groups are equivalent because they share a template.

Set the review conditions in advance

Choose a first review date and explain why it is appropriate for the question. Google's guidance says changes may take hours to months to appear in search. That range is a reason to avoid a universal seven-day verdict. The plan should account for the kind of change, the available observations and the possibility that the first review will be inconclusive.

Decide what would cause the team to pause or restart the record. A major site migration, a second edit to the same pages or a measurement failure could make the original comparison difficult to use. Add those events to the timeline rather than silently changing the plan. If the question changes substantially, create a new record and link it to the earlier one.

Name the person responsible for the review. A calendar reminder is more useful when someone owns the decision and knows which evidence to bring. The reviewer should be able to choose among continuing observation, repeating the exercise, keeping the change for a stated reason or reverting it. None of those outcomes requires pretending that the evidence is stronger than it is.

A worked record in plain language

The hypothetical record begins with twelve category URLs and a saved copy of each introduction. The proposed change is a shorter opening paragraph that explains the differences among three product groups. The hypothesis concerns whether that explanation supports a clearer path to the appropriate subcategory. The baseline export uses a named date range and filters, and the record identifies who captured it.

On release day, the owner adds the deployment time and confirms which pages received the edit. Three days later, another team member records a promotion affecting four of the categories. At the scheduled review, the group sees a change in the recorded measures but also sees that the promotion complicates interpretation. The conclusion is limited: the release is documented, the observation is real within the saved view, and the cause remains uncertain.

The next action could be to keep the clearer copy for its user-facing value while planning a narrower follow-up. That decision should say why the copy remains and what the team still wants to learn. It should not be rewritten as a successful ranking test. An honest record of uncertainty is useful to the next person planning a similar change.

Finish with a reusable handoff

Before editing a page, ask a colleague to read the record and point out anything they could not reproduce. Can they find the URLs, identify the original state, understand the proposed change and repeat the data view? If not, repair those gaps now. The review is usually easier while the plan is still small.

Afterward, retain the original hypothesis beside the final observation and decision. Do not replace the initial wording with a cleaner story that fits the result. A short archive of understandable records can reveal where a team's questions are too broad, where evidence repeatedly goes missing and which decisions deserve a better method. Start with one page change and a record that another person can actually follow.