# What Makes a Case Study's Results Believable? By AI Visibility editorial team AI-assisted article. A believable case study lets the reader reconstruct its main result and understand how much of the change the case can reasonably attribute to the work. Start by finding the metric, the people or events it counts, the comparison period and the evidence behind the claim. Then separate the result observed during the project from any estimate of what the project caused. Use this approach when evaluating a supplier or reviewing your own published case. A result can be useful for understanding what happened in one setting. A stronger claim about what would happen for another business needs evidence about the conditions that made the result possible. Read the method before treating a headline figure as a benchmark. Disclosure. AI Visibility is published by Lunasol, which offers GEO and AEO services. Two examples below are from the publisher or its owner. The enquiry calculation is fictional. About AI Visibility (https://aivisibility.ae/about/) Identify the result and what was counted Rewrite the headline as a complete sentence. Name the result, the unit counted and the period. “Enquiries rose” could refer to submitted forms, unique people, qualified opportunities or calls logged by staff. An apparently simple count depends on decisions about duplicates, cancelled requests, incomplete records and what qualifies for inclusion. For a rate, look for both the numerator and denominator. If a case says that 25% of requests qualified, it should say what counts as a qualified request and which requests make up the denominator. For a count, ask whether the scope remained stable. A business that started logging WhatsApp enquiries midway through a project can show a larger total even if the underlying number of people asking for help stayed stable. A public example comes from AI Visibility's 90-day plan. It reports that the second asks of repeated Google AI Mode questions retained 24 of the 31 websites counted on the first asks. It describes these as counts summed across four sectors. The figure is about retention under those runs, with the sector-level counting rule intact. A reader evaluating it should carry that rule alongside the fraction. AI Visibility's 90-day plan (/article/what-should-a-uae-business-do-in-its-first-90-days-of-ai-search-work-a-plan-week-by-week/) Check the baseline and align the windows Find the dates and what started each measurement window. Look for comparable geography, channels, staffing and definitions. A period that includes an opening, holiday or major campaign may have different conditions. If a case compares periods of unequal length, totals can hide an important difference. Monthly or weekly rates can add context if the source explains the period and the calculation. The baseline needs to answer a practical question. If the reader is deciding whether a process improved, they need a reference point from before the relevant change or another defensible comparator. A result during a programme can establish the volume recorded in that period. A before-and-after result needs both windows. An estimate of additional outcomes needs a basis for estimating what would have happened without the change. Lunasol's public DP Business Solutions case is a useful reading exercise. The case headline describes a 24-month period. Its period table labels the three sections as seven, twelve and six months, which add to 25 labelled calendar months. June 2024 to June 2026 can also describe a 24-month elapsed span between corresponding dates. Ask which boundary dates and partial months were used before dividing the headline count by 24 or treating the rows as equal exposures. The visible text gives a question to resolve. The underlying dates and counting records would determine the answer. Lunasol's DP Business Solutions case study (https://lunasol.ae/case-study/dpbusinesssolutions) Match the evidence to the claim An image can document a visible activity. A source export can support a count when its filters and method are known. A short comment can explain what a person experienced. Match each item to the claim it supports and check where you would need another source. Dubai Fitness Challenge's official 2 November 2026 Instagram reel looks back at Dubai Ride. A previously verified frame shows cyclists. The organiser's FAQ places the event on 1 November 2026. Those details give context about the activity and the sequence of event and recap. A claim about how many people participated would need a count and a definition, such as which participants were included. A claim about an increase would also need a comparable prior period. Official Dubai Fitness Challenge reel (https://www.instagram.com/dubaifitnesschallenge/reel/DfO6M6STOeg/), Dubai Fitness Challenge FAQ (https://www.dubaifitnesschallenge.com/en/faqs) Review screenshots and charts for the same limits. Ask which account, dates and filters they represent. Check whether an axis shows totals, cumulative values or a rate. If a case shows only a portion of the record, ask how it was selected. An image with a date can support the timeline while leaving the underlying definition or calculation open. View the selected official Instagram post (https://www.instagram.com/dubaifitnesschallenge/reel/DfO6M6STOeg/) Dubai Fitness Challenge’s official 2 November 2026 reel looks back at Dubai Ride. Separate observation, contribution and causation Describe what changed and what happened during the same period. If a company introduced a new enquiry form, changed its advertising spend and trained the sales team within the same month, several changes could plausibly explain a result. A case can show how each action was expected to work and which evidence supports that explanation. Keep the conclusion within what the design can distinguish. UK government evaluation guidance in the Magenta Book discusses approaches including contribution analysis and considering alternative explanations for observed outcomes. A case-study reader can apply that principle by asking what else changed and whether the source investigated it. If the article reports a comparison group, check how it was selected and whether it faced similar conditions. If it reports a mechanism, look for evidence that the expected intermediate steps actually occurred. Magenta Book Annex A on evaluation methods (https://www.gov.uk/government/publications/the-magenta-book/magenta-book-annex-a-analytical-methods-for-use-within-an-evaluation-html) Lunasol's published case describes several elements operating together, including advertising, content, funnels, email, CRM and sales processes. The headline figures are the publisher's account of the overall programme. A claim about one element's incremental effect would need evidence that isolates it or otherwise supports an attribution to it. Read the result at the level of the intervention the case actually describes. Lunasol's DP Business Solutions case study (https://lunasol.ae/case-study/dpbusinesssolutions) Recalculate a result before using it in a decision Imagine a fictional provider reports that its number of qualified enquiries rose after it revised a service page. The counts are 30 before and 50 after. The total number of recorded enquiries was 100 before and 200 after. Both statements can be true. The qualified count increased while the share that qualified fell. Measure Before After Change Recorded enquiries 100 200 100 more Qualified enquiries 30 50 20 more Share qualified 30% 25% 5 percentage points lower The qualified count rose by about 67% relative to the original 30. The qualification rate moved in the opposite direction. A business evaluating sales capacity may care about the larger number of qualified leads. A team trying to reduce time spent on unsuitable enquiries may care about the changing mix and the increase from 70 to 150 enquiries that did not qualify. Both interpretations depend on the definitions staying consistent. Suppose the business started advertising in a new channel at the same time. Ask whether the additional enquiries came from that channel, how its qualification rate compared and whether the change remained visible among existing sources. If the definitions changed, recalculate on a consistent basis where possible. The case could still report the total change while describing the channel expansion as an unresolved alternative explanation. When the result is small, look at the underlying count as well as the percentage. A change from two bookings to four is a doubling and an increase of two. An owner deciding whether to replicate the intervention needs enough history to know how much those counts normally move. Record the range of prior results if the case supplies it and distinguish ordinary variation from a persistent change. Check when outcomes become observable A form submission can be counted immediately. A qualified opportunity may require a call, and a sale may arrive weeks later. Check whether the case groups outcomes by the date of the original request or by the date the outcome was recorded. Those choices can answer different questions. Counting sales completed in October tells the reader what closed that month. Counting later sales from October enquiries tells the reader what happened to that group of enquiries. Consider the fictional example above. If the later period ended yesterday and some of its 200 enquiries are still being assessed, part of the apparent decline in qualification rate could reflect incomplete follow-up. Ask how the case handles unresolved records and whether both groups had the same opportunity to mature. An early snapshot can still describe the current record if it reports the pending cases and the date of the extract. A final comparison should explain how the groups were followed and when the outcome was considered settled. Reconcile the public result with the stated accounting basis. Revenue from orders, revenue collected and profit describe different things. A table may show a total programme result alongside spending on one channel. Dividing those figures can reproduce a published ratio while leaving a separate question about which sales were attributed to that channel. Ask for the numerator's scope and the allocation rule before using the ratio to compare campaigns. Keep the arithmetic check and the attribution check separate in the review record. Decide how much weight to give the result Case study result review record Headline claim, outcome and unit counted Data source, inclusion rules and exclusions Numerator and denominator where relevant Baseline and comparison dates, with partial periods identified Changes made and other events in the same period Evidence supporting observation, contribution or causal attribution Limits on transfer to your business and unresolved questions Use the record to decide what the case is good for. You might borrow a workflow because the intervention is well described. You might use the result as a planning scenario if the unit and timeframe are clear. Before using it as a forecast, check whether the source business had similar demand, budget, capacity and operating constraints. A case selected for publication can provide a detailed example without establishing how often similar results occur. Request the smallest clarification that would change your decision. If you are deciding between suppliers, it might be how they defined a qualified enquiry. If you are auditing your own case, it might be the exact dates behind a rounded period. A clear answer can resolve a material doubt without sharing a full private dataset. If the answer is unavailable, record the uncertainty and keep the claim at the level the public material supports. Read the headline again after reviewing the evidence. Check whether someone seeing only that sentence would understand the result, its period and the source of the claim. The case is easier to use when the table, narrative and summary describe the same measurement. The reader can then decide which lesson applies and what evidence they would need before acting on it.