Blog / Case studies /

A technical investigation written as a useful case study

A technical investigation can make a useful public case study even when the outcome is mixed. The hard part is writing up the reasoning without turning a hypothesis into a success story, or exposing the client.

A credible case study explains the question, the evidence, the work done and the limits of what you can conclude. It teaches readers how to think through a problem. It doesn’t need private details or invented outcomes to do that.

Confirm what can be shared

Start with permission and confidentiality. Determine whether the organisation, project details, screenshots, data and timing can be discussed. Removing a name may not be enough. A mix of niche, dates, locations and technical clues can still identify the client.

Agree the scope with the relevant owner before drafting. If permission is limited, generalise or leave details out. Don’t include customer data, account screenshots, proprietary code or internal documents without explicit approval. Anonymity isn’t a shortcut around permission.

I keep a private evidence record apart from the public draft. It holds the sources, dates and calculation methods a reviewer needs. The published story should reveal only what the owner authorised and what the reader needs to understand the method.

Frame the problem precisely

Explain the observed issue in neutral language. “Organic traffic collapsed” is a conclusion. A more useful description says which pages or signals changed, over what verified period and in which dataset. Skip the drama unless the evidence supports it.

State the question the investigation sought to answer. Was the team checking a technical implementation, comparing page groups or validating a proposed change? A focused question helps the reader follow the work and prevents the story from becoming a list of unrelated SEO tasks.

Distinguish symptoms from causes. A crawl finding may show a configuration. It doesn’t prove that configuration caused a traffic change. Explain what evidence linked the observation to the proposed cause, and what other explanations remained possible.

Show the method, not confidential material

Describe the diagnostic sequence at a level another practitioner can learn from: which systems were examined, what patterns were compared and how the team ruled out alternatives. Don’t publish sensitive access details or a client’s exact URL set without permission.

Use a small table or diagram if it makes the process easier to understand. Label illustrative examples as illustrative. Never manufacture before-and-after numbers, anonymised quotations or representative screenshots. If you can’t share exact data, say so and focus on the reasoning.

Include the changes actually made and the verification performed. If no implementation happened, say the investigation ended with a recommendation. Don’t imply a proposed fix produced a result nobody measured.

Report outcomes with restraint

If results are available and approved, define the metric, comparison period and source. Say whether the measure is observational or experimental. Don’t cherry-pick a favourable window, and don’t call correlation causation.

A technical fix can be successful even when a ranking outcome is uncertain. For example, the team may confirm that an intended page is now crawlable; that does not guarantee traffic growth. State the outcome at the level the evidence supports.

If the result was inconclusive, explain what prevented a stronger conclusion: limited data, concurrent changes, short follow-up or an unresolved dependency. That is not a weakness in the case study. It helps readers avoid repeating an overconfident interpretation.

Extract lessons that travel

Separate transferable principles from client-specific choices. A particular site’s architecture may not generalise, while the investigation method—checking rendered output, comparing templates or validating redirects—may help others.

Describe constraints that shaped the work, such as platform limits or available access, only when they can be shared. Do not imply that every site should adopt the same tool or sequence. Give readers questions to ask in their own context.

Have a technical reviewer and an appropriate business owner check the draft. Ask them to verify facts, confidentiality and whether the stated conclusion follows from the evidence. Keep a record of approval and any redactions.

Make the story useful without a dramatic result

A case study does not need a traffic graph to be valuable. It can show how a team formed a diagnosis, compared implementation options and verified a technical correction. Be precise about the deliverable: “the canonical output matched the approved rule” is a different claim from “search performance improved”. Readers can learn from the former without being promised the latter. I’d rather publish that than a graph I can’t defend.

Use a simple evidence table with columns for question, source, observation and limitation. This makes gaps visible and helps reviewers catch a leap in reasoning. If evidence comes from a tool, mention its role and date where that matters, but don’t imply the tool proved causation by itself.

Include one decision the team chose not to make and why, if permission allows. Constraints and rejected options often explain the work better than a polished success story.

Make review part of the evidence chain

Have a reviewer who was not involved in the investigation read the explanation and ask them to trace each conclusion back to a source. If they cannot tell which observation supports a statement, revise the wording or narrow the claim. This simple check catches leaps the author stopped noticing long ago.

Keep an internal note of what cannot be published and why, so later edits do not reintroduce restricted details. If the article is updated after publication, retain the version history and correct a claim visibly when the change affects the reader’s understanding.

A useful case study can end with open questions and the next measurement needed, which gives readers a realistic sense of the work’s boundary. What conclusion in your draft would you be least comfortable defending if a reader asked to see its evidence?