Direct answer: A percentage is only interpretable with the starting count, ending count, time window, definition of the measure, and source. If those facts are unavailable or permission is missing, describe the work and observable change without a numeric performance claim.

“Leads increased 40% after the redesign” sounds precise. It might mean 10 form submissions became 14 in a month; it might mean 1,000 became 1,400 across a quarter; it might include spam or calls that were not tracked before. The arithmetic is the same, but the business interpretation is not. A reader needs enough context to see what the number means and whether the redesign plausibly explains it.

Build a claim card before writing the headline

For each proposed result, record: measure, starting and ending counts, denominator, dates and comparable periods, data source, scope, known changes, and approval to publish. For example:

Qualified website inquiries, as classified by the sales team, rose from 25 in April to 35 in May 2026 (40%). The comparison covers one location. The site redesign launched on May 1; the business also increased paid-search spend during May. CRM export reviewed June 5, with the client approving these aggregate figures for publication.

That wording is longer than “40% more leads,” but it lets a buyer distinguish a recorded increase from a causal estimate. It also gives the business a way to correct the public claim if the definition or source changes. The example above is illustrative, not an ARCHV client result.

Google's generative AI guidance emphasizes original, useful information and warns against assuming special markup is needed for AI search. A case study's original value is the specific problem, decisions, constraints, and evidence a buyer cannot get from a generic advice page. Giving an AI system a clean paragraph to quote is a useful editorial by-product; it cannot guarantee a citation or a correct summary.

There is a reason to inspect summaries rather than trust a displayed source link. A 2023 peer-reviewed human evaluation by Liu, Zhang, and Liang audited four generative search engines then available and found that some citations did not support the generated sentences. Those historical results do not measure today's systems or prove that any particular case study will be misstated. They support a modest practice: if an AI answer cites your case study, check whether its claim matches the cited text and correct the underlying page if your own wording is ambiguous.

Show the work behind the number

Explain what changed: the old inquiry path, the research or usability findings, the new page or form, and the launch date. State what the team controlled and what it did not. If campaign spend, pricing, seasonality, or sales staffing changed during the measurement window, name it. Avoid a before/after layout that implies the website caused every subsequent outcome. If the sample is small, present counts before percentages.

The existing guide to writing case studies that attract fitting clients focuses on which stories and decisions to show. This article is the evidence check for a numerical outcome. A separate campaign attribution article explains when a causal claim needs a stronger design. ARCHV can help shape a scoped project's story and website presentation; the client must supply and approve the underlying performance data. Start with a redacted source record and the claim you want to make in a website estimate.