How to Check Whether an Agency's Results Claims Are Real

How can you tell if an agency's results claims are real?

Ask seven questions of any results figure: what was counted, against what baseline, over what window, measured how, by whom, whether the same number appears on every page the seller controls, and who produced it. A figure that answers all seven is checkable; one that answers fewer than three is a number you cannot yet check, however large. The test takes about a minute per claim, needs no login, and as of 5 October 2026 our own homepage still fails it.

What the test found, at a glance:

• Unsourced persuasion: in the fourth of four Advids press releases, dated 2026-08-25, the buyer-behaviour figures 37%, 75%, 53% and 70% name no source.
• Fresh dates, old pages: on 2026-08-26 TubeScience's sitemap carried all 13 of its URLs re-stamped to the same millisecond, with no content change.
• A near-pass: NewForm's method note answers five of the seven questions in two sentences, though its own headline counts the ads differently.
• Our own pages: four Dream Aspect homepage figures are hidden, undated and unscoped, and one engagement figure reads 58.31% on one page and 58.41% on another.

What are the seven questions to ask about any results claim?

A results claim is any figure a seller publishes about an outcome it says its work produced: views, engagement growth, return on ad spend or a count of campaigns. Provenance is the record of who produced that figure and how, and it is what the seventh question asks for. The seven questions cover what a figure measures, where it came from and whether the seller tells the same story everywhere. Each has a plain pass and a plain fail, so you can score an agency's results page, a pitch deck or a monthly report without any access to its data. With only a minute, stop at the first three: a metric, a window, and a baseline or a count.

1. Metric: what was counted? It passes with a defined platform metric and fails with vague words such as results or growth, or no unit at all.
2. Baseline: compared with what? It passes with “From X” or “against a control” and fails with a percentage that has no comparator.
3. Window: over what period? It passes with a stated period, such as a year or a number of days, and fails with no period or only an “updated” stamp.
4. Method: how was it counted? It passes with a sentence you could hand to an analyst and fails with silence.
5. Measurer: who measured it? It passes when it names the agency, the platform, the client or a third party, and fails when a result was simply achieved.
6. Consistency: is it the same number everywhere? It passes when the figure is identical on every page the seller controls and fails with two figures for one result.
7. Provenance: who produced the figure? It passes with a sentence that attributes it and fails with a figure that floats.

Do the numbers that persuade you carry a source?

Not in the agency press release we read figure by figure. Advids, a B2B video production agency that competes with Dream Aspect, issued four vertical releases on EIN Presswire between 2026-08-05 and 2026-08-25, one for each of these markets:

• AI/ML
• HealthTech
• CleanTech
• IT Infrastructure

In the fourth, dated 2026-08-25 and titled Advids Launches Dedicated Video Production Solution for IT Infrastructure as 70% of Engineers Research Brands on YouTube, the market-context figures name their source: the headline 70% is attributed to the State of Marketing to Engineers study from TREW Marketing and GlobalSpec, and a 91% figure to Brightcove. The buyer-behaviour percentages that do the persuading, 37%, 75%, 53% and a further 70%, are attributed to nobody.

That is not a claim that the unsourced figures are wrong, only that you cannot check them. We read one of the four Advids releases on EIN Presswire figure by figure, so the finding covers that release, not the agency's output as a whole. A press release is seller-authored copy, and an unsourced figure there is exactly as checkable as one on a homepage. The quickest version of the whole test is the last question alone: read any results page asking who produced each figure, and see what is left.

Is a recent “updated” date evidence of a fresh result?

No: a date on a page is metadata, and metadata can be rewritten in bulk. TubeScience's sitemap, read on 2026-08-26, carried all 13 of its URLs re-stamped to a single value, 2026-08-20T20:09:24.608Z, with no change to any page. Its machine-facing llms.txt file still reads "Last updated: 2026-03-04", as it did again when we re-read it on 5 October 2026. On that same day, 13 URLs in TubeScience's sitemap again shared one identical timestamp, a newer one.

A bulk re-stamp of the TubeScience sitemap can be a platform artefact rather than anyone's decision. We do not know which this was, TubeScience competes with Dream Aspect, and we are describing what a date can tell you, not anyone's intent. The lesson for a buyer is narrower than the example: a recent “updated” stamp tells you when a file was touched, not when the result behind it was measured, so ask for the window every time.

Where else does a seller state its numbers, and do they agree?

An agency states its numbers on at least three surfaces, and the same-number check compares them digit for digit wherever a figure appears twice. In Dream Aspect's August 2026 sweep, eight of the eleven competitors published the third surface, a machine-facing text file at /llms.txt. The three surfaces to compare are these:

• the homepage, including figures in its markup
• the case study, pitch deck or monthly report
• the llms.txt file, written for AI search engines rather than people

A discrepancy between two surfaces may be nothing worse than a transcription slip, but the seller cannot say which figure is right without checking, and how they answer tells you how the number was produced. The same check works on a results guarantee: the number promised in the contract should match the number on the website.

Do not expect an outside referee to have run the check for you. Of the 20 competitor domains Dream Aspect profiled for the Los Angeles and San Francisco markets on 2026-08-21, none has a Wikipedia article, so there is no encyclopaedia entry to compare their figures against. Ask the agency which figure it will defend in writing.

What does a passing claim look like?

A passing claim is a sentence that carries its own audit trail. NewForm, a competitor of Dream Aspect, publishes on its NewForm Perspectives page Hook Rates Are a Scam: "Dataset: 4,067 ads across 23 NewForm client accounts, tested through Framework between Q4 2024 and Q1 2026." and "Winners declared at 95% statistical confidence against batch control on downstream conversion outcomes." Those two sentences answer metric, baseline, window, method and measurer, five of the seven questions, and both still read word for word on the page as of 5 October 2026.

Run the same-number check on the NewForm hook-rate page, though, and the NewForm headline and summary state a larger ad count across the same 23 accounts than the dataset line does. It may be a different cut of the data, but the page does not say which count the 95% result rests on, so that is the question to ask. A strong method note still has to agree with the headline above it.

Darkroom, an advertising and growth marketing agency headquartered in New York, shows the last question, provenance, done right. Its machine-facing llms.txt file carries: "Provenance note: all figures above are Darkroom first-party data reported by the agency unless otherwise attributed." We did not audit Darkroom's figures; the note does something smaller and more valuable than being right, which is to tell every reader whose number they are about to repeat. Copy the note, not the tone.

Do our own numbers pass this test?

No, as of 5 October 2026. Dream Aspect's homepage markup carries four aggregate figures in a section served to your browser but styled not to display, so a visitor never sees them:

• a count of brands
• a count of campaigns
• a 20M total-engagement figure
• a $1M active-ad-spend figure

On the Dream Aspect homepage those four figures carry no period, no definition of “engagement” and no measurement basis. The reporting-tile figures exist in the homepage HTML only as image alt text, where engagement growth reads 58.41%, while our wellness-brand case study states 58.31% for the same Instagram result.

Stated properly, the case-study figure is: +58.31% Instagram engagement growth, period 2025, measured by us, self-reported, no control group, method not stated on the page. We quote the lower published number because the higher one flatters us, and the 0.10-point gap is unresolved. Our headline view-growth figure is left out: as published it carries no unit, and we will not restate it until it does. Whoever wrote this test also benefits if you skip Dream Aspect's own scorecard, so until our homepage carries each figure with its window, definition and measurer, treat our aggregates as unverified, the verdict this article hands everyone else.

Want to run this on our numbers?

Run the seven questions on Dream Aspect's case study and homepage before a call; we would rather you raised the gaps than found them later. Our piece on the format strategy is the closest we come to saying how the case-study figure was counted. Still building the shortlist? How to choose a content agency in Los Angeles covers the questions that come first, and social media engagement rate benchmarks for 2026 give you a yardstick for any engagement figure an agency quotes.

Want to see Dream Aspect's own prices before you compare agencies? The Dream Aspect pricing calculator shows the price on the page: choose a one-off shoot or a monthly plan, then add or remove deliverables such as a shoot brief, priority delivery, CGI or UGC.

Build your quote on the pricing page →

Prefer to talk it through? Talk to us → or book a 15-minute call from the pricing page.