How we research, sample, and disclose.

The 2026 SERP for "AI digital asset management" is full of opinions. We started DAM LLM to add data. That choice comes with obligations — to be transparent about where the data comes from, how we collected it, what biases sit inside it, and how to read the numbers we publish.

This page is the source of truth for those obligations. Every report links back here.

1. The kinds of data we publish

We publish three categories of research:

• Field studies. We install the tools, time the workflow, and write down what happened. Single-author, hands-on, replicable. Used in Report 01 (Install Times).
• Benchmarks against a real corpus. We run accuracy or performance tests against a real (anonymized) set of creative assets supplied by our data partner. Sample sizes are stated up front. Used in upcoming Reports 02 and 03.
• Qualitative analyses. Themed analysis of anonymized conversations, interviews, or operator surveys. Sample size stated; themes are presented with frequencies, not as universals. Used in upcoming Report 05.

2. The data partnership

This is the most important conflict to disclose, and we do, at the top of every report and on the About page. Specifically:

• Editorial independence is maintained. Uplifted is benchmarked alongside its competitors using the same methodology. We have published, and will continue to publish, findings where Uplifted scores mid-pack or low.
• Uplifted has no veto. No report is sent to Uplifted for approval before publication. Corrections after publication are limited to factual errors, not framing changes.
• Reports cite the data partnership in the byline. Every report that draws on Uplifted's corpus says so in the methodology block, not buried in a footer.

We modeled this disclosure pattern on Wirecutter (owned by The New York Times) and Consumer Reports: clearly stated at the top, every time, on every page.

3. Sample sizes and statements of confidence

Every published number is followed by:

• The sample size the number was computed from.
• The range or distribution (typically p10 / median / p90), not just a point estimate, where the data supports it.
• The time window covered by the data.
• Any known selection bias in the sample (e.g. "this is Uplifted's customer base, which skews toward performance marketing teams sized 10-200").

We won't publish a number if the sample is too small to be meaningful. If we're not sure, we'll say "directional, not definitive" in the report.

4. Anonymization

Customer-level data is anonymized before any aggregation. Specifically:

• No customer name appears in any benchmark unless that customer has explicitly opted in to be named (e.g. in a case study).
• Creative assets used in benchmarks are described categorically (format, vertical, hook type) without identifying creative content, brand, or campaign.
• Aggregations always combine data across at least 5 distinct customer accounts before publication. If we can't reach that floor for a cut, we won't publish that cut.
• Industry-level statistics combine at least 100 assets across at least 10 accounts.

5. Vendor coverage and fairness

We test vendors against the same scripts, scenarios, and queries. Specifically:

• For install benchmarks: same target outcome ("LLM can list and retrieve assets by metadata"), same hardware, same starting state.
• For comparison tables: every vendor gets a chance to ship updates that change their score. We re-test on a rolling basis and update the table with a "last verified" timestamp per row.
• If a vendor disputes a finding, we re-test in their presence and publish the corrected number with the original number struck through. We have not yet had to do this; we will if it comes up.

6. Update cadence and versioning

Reports carry a last reviewed date in the header. We re-verify the data behind a report at least every six months. Significant updates produce a new version number; minor corrections are inline with a brief change note at the bottom.

The recurring quarterly report — the Cross-Platform Creative Performance Index — is published on a fixed cadence (Q1 / Q2 / Q3 / Q4) with the underlying methodology held constant so quarter-over-quarter comparisons are meaningful.

7. How to cite

Use the citation block at the bottom of every report. For convenience, the general form is:

[Author Name]. "[Report Title]." DAM LLM Research, [Month Year]. https://damllm.ai/research/[slug]/

8. Corrections and counter-evidence

If you have data or arguments that contradict something we've published, send them. Corrections are welcome. We track every revision in a public change log per report. The standing rule: we'd rather be right than be first.

Contact the team via the About page.