Ask a marketing team where a number in their case study came from and you get one of two answers. Either someone opens a folder and shows you the dated export, or there is a pause. The pause is the problem this article solves.
A marketing proof library is the organized store of first-party evidence behind everything you publish — SEO results, PPC performance, case study claims, and the statements answer engines might repeat on your behalf. It is unglamorous infrastructure, and it is now the difference between content that survives scrutiny and content that quietly cannot be defended.
The three evidence layers
Most evidence problems come from collapsing three distinct things into one.
Layer 1: raw source
The untouched export or screenshot from the system of record — Search Console, Google Ads, GA4, a CRM, an invoice, a platform dashboard. Stored as retrieved, with the retrieval date and the exact date window it covers. Nothing edited, nothing summarised.
Layer 2: documented interpretation
What the raw data means, written down: the window, the segment, the filters applied, the comparison baseline, the known caveats, and who did the analysis. This is where most disputes are actually resolved, because disagreements are usually about method rather than numbers.
Layer 3: published claim
The sentence that reaches the public — a headline stat, a case study result, a homepage number. Every published claim carries the ID of the interpretation that supports it. If it has no ID, it does not ship.
The 14 proof-record fields
- Record ID — stable, referenced by any content that uses the evidence.
- Claim supported — the specific statement this record backs.
- Source system — the platform or document the data came from.
- Retrieval date — when the data was pulled.
- Date window covered — explicit start and end dates.
- Segment and filters — property, account, campaign, country, device, query filters.
- Metric definitions — what each figure counts, in the platform's own terms.
- Method notes — calculations, exclusions, baselines and rounding.
- Known limitations — sampling, thresholding, attribution windows, data gaps.
- Client or subject — whose data it is.
- Consent and approval — what was permitted, by whom, in writing, and when.
- Ownership and rights — who owns the asset and where it may appear.
- Retention period — when the record is reviewed or deleted.
- Reviewer and review date — who verified it and when it next needs checking.
Source controls by channel
SEO evidence
Search Console is the system of record for organic impressions, clicks, CTR and average position — and its quirks belong in the record. Average position is an average of best positions, query data is thresholded, and totals differ between property types. Always store the date window: our own property, for instance, recorded 64 clicks and 127,836 impressions at 0.0501% CTR and average position 31.42 for July 25 to August 21, 2026, with aeo services at 159 impressions and average position 16.68, and seo services with results at 76 impressions and average position 43.89. Those are visibility observations for one site in one window — not search volume, difficulty or market estimates, none of which Search Console reports.
PPC evidence
Ads data needs its conversion action definitions, attribution model, lookback window and any value rules captured with it, because a "conversion" changes meaning silently when settings change. Store spend in the account currency and note conversion rates used. Screenshots of the settings that produced the numbers age far better than the numbers alone. Our PPC advertising reporting practice treats settings baselines as part of the evidence, not context.
AI search evidence
There is no authoritative report for how AI experiences use your content. What you can hold is dated, reproducible observation: the prompt used, the surface, the date, a screenshot, and whether your content was cited or paraphrased. Label it as observation and never present it as a measured share of anything. This is the discipline behind our GEO services.
The seven-step workflow
- Trigger. Any new claim, case study, report or content brief opens an evidence request.
- Retrieve. Pull raw data from the system of record, unedited, with the window fixed.
- Redact. Remove personal data, account identifiers and anything not covered by consent.
- Interpret. Write the method notes and limitations while the data is still fresh.
- Approve. Obtain written client permission for anything client-specific.
- Publish with reference. Attach the record ID to the content that uses it.
- Review. Re-check on a schedule; retire or update claims whose evidence has expired.
Privacy and consent
Evidence libraries accumulate personal data quickly — names in CRM exports, emails in call logs, faces in site photography. Three rules keep this manageable. Collect only what supports the claim. Redact before storage rather than before publication. Set a retention period per record type and actually honour it. Written consent should state what may be named, what must be anonymised, which metrics may be disclosed, and for how long.
Using the library in content
The library changes how content gets written. Case studies cite windows and methods instead of round numbers with no provenance, which is the standard we hold across our portfolio. Service pages make narrower, checkable claims. Client reporting separates measured results from observed ones, which is the backbone of our marketing reporting. And organic content becomes materially harder to dismiss, which is the practical goal of SEO services in a market saturated with unverifiable copy.
A small practical habit helps here: keep published claims tight and readable. Running draft copy through a word counter before publishing catches the padded paragraphs where unsupported claims tend to hide.
The minimum viable library
You do not need software. Create a dated folder per client and per channel, a single spreadsheet indexing the 14 fields, and one non-negotiable rule: no public claim without a record ID. Backfill the evidence for your five most-cited existing claims first — those are the ones most likely to be challenged. Everything after that is refinement.
See what your current claims look like from the outside
Before you rebuild the evidence behind your content, it is worth seeing how that content currently reads to search engines and answer engines.
Proof libraries are boring to build and impossible to fake retroactively. The teams that start one now are the ones whose numbers still hold up when someone finally asks where they came from.
