A case study becomes useful to buyers and answer systems when its central result can be repeated without becoming larger, cleaner or more certain than the project record allows. Baseline, intervention, observation period, result, source role and material limitation must travel together.
If lifting one sentence out of the page turns a bounded observation into a universal promise, the case study is not ready.
TL;DR
- Write the result sentence before the victory story, and let every bracket in it be a decision the record must fill.
- Build the evidence record first: subject, baseline, intervention, period, result, source role, other changes, limitation, permission, review trigger.
- Every percentage needs its denominator; every source needs its role named beside the claim.
- Run the extractability test: paste the strongest paragraph into a blank document and see what survives.
That is the test most victory stories avoid. The graph looks good. The client is pleased. Sales wants the page before the next meeting. Then the draft says the work transformed performance and quietly loses what changed, when it changed, what was measured and what else happened during the period.
Everybody approving the page may know the missing context. A buyer arriving six months later does not. Neither does an AI system extracting one convenient passage. My rule is simple: write the result so it can survive the quote.
Write the result sentence before the victory story
Start with the proposition the business is prepared to have repeated elsewhere. Do this before writing the chronology, testimonial or celebratory introduction. Use an unfilled structure rather than inventing a polished claim:
For[defined subject or page set],[defined measure]changed from[baseline]to[observed result]between[start]and[end], after[bounded intervention]. The figures come from[named source and scope].[Material limitation or competing explanation].
Every bracket is a decision. If the record cannot fill one honestly, the sentence should become narrower or remain unpublished. A missing baseline cannot be repaired with an adjective. An unknown period cannot be hidden inside after the campaign.
I would rather publish a modest result with clean boundaries than a heroic number nobody can defend. The smaller claim is often more persuasive because a serious buyer can see exactly where it begins and ends.
Build the evidence record before the public page
The case study needs a working evidence record behind it. This is not public copy or a marketing mood board. It is the material an editor, client and future reviewer can use to reconstruct the proposition.
| Field | What the record must establish | Common failure |
|---|---|---|
| Subject | Exact client, account, site, page group, market or campaign | The result quietly becomes company-wide |
| Baseline | Measure, value, unit, source and prior period | One convenient low point is presented as normal |
| Intervention | What changed, where, when and by whom | A whole retainer gets credit for one movement |
| Observation period | Start, end and comparison window | “Within months” replaces dates |
| Result | Value, unit, denominator and calculation | A percentage appears without quantities |
| Source role | First-party system, client record, customer report or independent source | A dashboard screenshot is called independent proof |
| Other changes | Material activity outside the claimed intervention | Rebrand, paid media or seasonality vanishes |
| Limitation | What the record cannot establish | Correlation becomes causation |
| Permission | What may be named, shown and quoted | Internal data becomes public through enthusiasm |
| Review trigger | What would make the claim stale or misleading | The result lives forever after conditions change |
The public prose is downstream of this record. It must not outrun it.
For relevant non-broadcast marketing communications, the UK CAP Code says marketers should hold documentary evidence before publishing objective claims capable of substantiation and present significant qualifications clearly (CAP Code, section 3). CAP’s substantiation guidance also warns that testimonials alone are unlikely to substantiate objective claims (CAP substantiation guidance). A happy quotation and a measured outcome are different evidence objects.
A baseline is more than the word before
A baseline is a defined observation, not the first number that makes the result look dramatic. Record the metric, source, unit, population and period. If the result concerns organic enquiries, specify what counted as an enquiry and how organic attribution was assigned. If it concerns revenue, state whether the figure is gross, net, attributed or a client-supplied band. If it concerns AI search optimisation, say whether the observation is a cited answer, mention, sampled response, click or something else. Then test comparability:
Not every difference invalidates the case study. It may simply narrow the claim. If new call tracking improved observation as well as performance, state that. If the company doubled available appointments, do not award the website sole credit for additional bookings. The baseline exists to make the result interpretable, not impressive.
Describe only the intervention that shipped
The intervention is the work reasonably connected to the observation. It is not every service listed on the invoice. Separate what was released from what was discussed, recommended or left pending. Name the affected page set, campaign, template, profile, tracking route or content asset. Record the implementation date and responsible party.
This distinction matters because case studies often credit strategy for changes that never shipped or award one channel the results of several. We improved digital growth is not an intervention. The company replaced the service-page template across the defined page set on the recorded release date is at least inspectable. Do not force causal language where the design cannot support it. Useful wording includes:
After the release, the recorded measure changed…During the observation period, the account recorded…The available records show an association between…The project included this work, while these other conditions also changed.
Those sentences are not timid. They are the difference between a case study and retrospective fan fiction.
Keep time attached to the outcome
Dates control meaning. A seven-day spike, twelve-month trend and year-on-year comparison answer different questions. State when the intervention went live, when measurement began and ended, and whether a lag or excluded period applies. If the business is seasonal, choose a comparison acknowledging it. If the observation is early, call it early rather than a settled outcome.
Avoid converting a temporary state into a permanent product claim. Recorded during the first 90 days after release is bounded. Delivers more leads is an ongoing promise. The review date matters too. A case study involving a retired offer, old platform, former product or changed market should not remain a current sales claim merely because the URL still works. Trigger review when the method, permission, underlying source or commercial offer changes.
Every percentage needs its denominator
Percentages create authority cheaply. The denominator is where that authority gets tested. For every material number, preserve what was counted; value and unit; denominator where applicable; date range; population or page set; calculation; source; and exclusions.
Enquiries increased by [percentage] is incomplete if the reader cannot tell whether the comparison is one month against twelve, forms against all contact routes, or raw leads against qualified enquiries.
Where confidentiality prevents exact figures, use an approved range, indexed change or qualitative result and label it honestly. Do not manufacture precision from redacted data. A disclosed band is more useful than a suspiciously exact number with no inspectable basis. Performance marketing should not make the denominator optional merely because the chart fits nicely into a slide.
Name the role of every source
A case study usually combines proof with different functions. Keep those roles visible. First-party platform data can show what a system recorded under its rules. A client CRM can show stages and outcomes recorded by the client. A customer quotation can show that person’s reported experience. An issuer can verify a credential or status within scope. The agency’s project log can show what work it performed and when. None automatically proves the complete story.
A screenshot is not independent because it contains numbers. A testimonial is not a performance dataset. A client statement remains client-supplied. Analytics cannot establish activity it failed to observe. Label the source role beside the claim rather than pouring everything into a decorative Results block. The evidence becomes more persuasive when a reader can see who produced it, what it covers and what it cannot prove.
Got a case study sales is impatient to publish? We rebuild the evidence record, write the result sentence and tell you which claim it can honestly carry.
Put the caveat where extraction cannot lose it
Do not hide the limitation in a disclaimer at the bottom. Keep it in the same passage as the result it controls. A material caveat might concern simultaneous website, branding, offer or pricing changes; seasonality or a changed comparison period; incomplete call, form or revenue tracking; a small sample; client-supplied data not independently audited; an outcome limited to one market, page set or service; inability to isolate one intervention’s contribution; or an observation that has not persisted.
The caveat should not contradict the headline after the headline has already sold the result. It defines the honest size of the claim from the start.
The extractability test: paste the strongest result paragraph into a blank document. Does it still name the subject, measure, period, source role and controlling limitation? If not, context is not attached tightly enough.
Make the method useful to the next buyer
A case study is not finished when it proves something happened. It should help the next reader decide whether the approach fits their situation. Explain the diagnosis, alternatives considered, chosen intervention and conditions that mattered. Show what the team checked, what it refused to assume and what depended on the client. Keep trade-offs and failed routes when they teach something real.
The useful lesson is rarely hire us and this number happens again. It is more often that this symptom was traced to this constraint; this evidence changed the priority; this intervention fit these conditions; this result should not be generalised beyond this scope; and a similar company should collect these records before choosing a fix.
That is non-commodity value. Another agency can imitate the headline. It cannot cheaply reproduce the project decision record behind it. This is where case studies support customer acquisition without becoming aggressive sales copy. They let a suitable buyer recognise the conditions, and an unsuitable buyer recognise the difference.
AI visibility does not create a special case-study formula
Google says ordinary Search foundations remain relevant to AI Overviews and AI Mode, with no special optimisation required. A page must meet normal technical eligibility to be considered as a supporting link, but eligibility does not guarantee crawling, indexing or display (Google AI-feature guidance).
An extraction-safe case study is therefore not a citation trick. It is a stronger source for any person or system trying to understand the business. Clear subjects, bounded claims, visible evidence and attached limitations reduce guessing. They do not create a right to inclusion. Google’s people-first guidance also asks whether material offers original reporting, research or analysis and whether sourcing and expertise are clear (Google helpful-content guidance). A reconstructed project record can provide that original contribution. Rephrasing a testimonial cannot.
If AI visibility matters, sample one fixed question under recorded conditions. Capture the product, mode, prompt, date, visible citations and exact answer passage. Compare the extracted statement with the case-study result sentence. Classify the observation: no answer, absent source, present without visible citation, visibly cited, accurately extracted or materially distorted. One sample proves only what appeared under those conditions. A cited page can still produce no visit or enquiry. Visibility, accurate extraction and business impact are three different observations.
Give the draft a real release decision
Every case study should leave editorial review in one state:
| Decision | Use it when |
|---|---|
| Publish | Claim, evidence, permission, scope and caveat are complete |
| Publish a narrower claim | The record supports less than the draft promises |
| Hold for evidence | A material number, source, period or intervention cannot be reconstructed |
| Hold for permission | Evidence exists but public use is not approved |
| Remove the result | Method or lesson is useful but outcome cannot be substantiated |
| Retire or annotate | The result was supportable but the offer, evidence or conditions changed |
Do not let needs a better introduction disguise cannot prove the result. Editorial polish happens after claim permission.
Before approval, extract the strongest passage and read it without the title, chart or surrounding story. Ask a person outside the project what they think it proves. If their interpretation is larger than the record, tighten the language. If the record is incomplete, hold it. If the passage survives, the case study has become more than a victory brochure: it is a result another person can inspect, quote and use without being misled.
Publish the result that survives the quote
Write the result sentence first. Build the evidence record. Define baseline and intervention. Keep time and denominator attached. Label every source. Place the caveat beside the outcome. Explain the method as a decision rather than a miracle.
I would rather lose a dramatic headline than publish a result sales cannot defend when a cautious buyer asks the second question. A good case study does not remove uncertainty. It controls uncertainty well enough that the result remains true after it leaves the page.
Good to know
FAQs
The defined subject or page set, the measure, the baseline, the observed result, the start and end dates, the bounded intervention, the named source and scope, and the material limitation or competing explanation. If the record cannot fill a bracket honestly, narrow the sentence or hold it.
No. CAP’s substantiation guidance warns that testimonials alone are unlikely to substantiate objective claims, and the CAP Code expects documentary evidence before publishing claims capable of substantiation. A quotation shows what that customer reported, not a measured outcome.
Only with the denominator, date range, population, calculation and source attached, or as an honestly labelled approved range or indexed change where confidentiality prevents exact figures. A percentage without a denominator cannot be interpreted.
It makes the page a stronger source for any person or system, but it is not a citation trick. Google says ordinary Search foundations apply to AI features and eligibility does not guarantee display. Visibility, accurate extraction and business impact remain three separate observations.
Next step
Would your headline result survive the second question?
We reconstruct the evidence record, write the result sentence and run the extractability test before a case study reaches sales.