Level: Optimize

B2B evidence: what can you really claim?

A claim, relevant evidence, an explicit check and a conclusion within scope.
Estimated reading time:
Verifiable evidence dossier: three bound sheets identify the source, date and verification.

Social proof can reassure, but what does a review, metric or case study actually support? Connect each claim to relevant evidence, state its limitations and decide what to publish—with three Edikka self-audits and a six-step method.

Reading criterion

Check a claim, with or without a number

Social proof—reviews, testimonials and client references—provides signals of trust. To assess what it supports, examine its provenance and scope. Verifiable evidence is material a reader can examine to assess a specific claim within a defined context. It may be quantitative, such as a measurement report, or qualitative, such as an accessible deliverable, a demonstration or an attributed testimonial. Its value depends on the conclusion it actually supports.

Apply this test to your page: can a reader find the evidence, understand what they are checking and identify the boundary of your claim? A review first establishes that someone expressed a judgment. A deliverable shows what was delivered. Neither, on its own, demonstrates an increase in sales.

This method is for teams writing B2B pages, case studies and commercial proposals. Start with the three self-audits, then document a review of one claim on your site. Treat this as a first review, not certification.

Reviewing our own content

Three self-audits: performance, conversion and a deliverable

We reviewed three claims published by Edikka and retained the evidence and observations from 11 October 2026. These cases illustrate the method; they are neither a representative sample nor independent validation. A new observation does not erase an earlier gap.

UTH: a historical result is not its measurement report

Claim reviewed — The earlier English edition stated: “100 mobile performance, 1.3 s mobile LCP and 0 CLS” for a Lighthouse measurement dated 8 June 2026.

Available evidence — The internal summary dated 8 June 2026 exists: it records Lighthouse CLI 9.6.8, two mobile runs and two desktop runs. The corresponding raw reports were not found in the supplied historical folder. Our initial review missed this summary; we are correcting that omission. The UTH project page now links to current observations and their limitations.

Check performed — The historical summary selects the worse of two runs only when it improves the previously published figure: an upward-only publication rule that does not document every possible regression. Our separate 11 October series of three Lighthouse 13.0.1 tests (Chrome 157, 412 × 823 mobile emulation, ×4 CPU slowdown) returns 99, 99 and 100, median LCP 1.18 s and zero CLS. A counter-check executed by Google PageSpeed returns 87 and LCP 3.7 s with Lighthouse 13.5.0.

What this supports — The summary documents what Edikka recorded in June. The three local JSON reports document their October executions. The Google PageSpeed report, retained despite its less favourable result, documents a separate run in a different environment.

What it does not prove — Without the raw June reports, the historical execution remains unverifiable with the available evidence. Local and Google versions and conditions differ: do not average their scores or infer a deterioration over time. PageSpeed reports insufficient CrUX data for the tested page; this establishes neither zero visitors nor absence of all origin-level records. None of these tests measures conversion.

Status and decision — Historical claim: not verifiable with the available evidence. Action: cite the summary and its selection rule, stop presenting the old figure as complete evidence, and display the new observations and limitations separately.

Revised wording — “On 11 October 2026, three local mobile tests returned 99, 99 and 100 with Lighthouse 13.0.1. The Google PageSpeed counter-check returned 87 with Lighthouse 13.5.0. Evidence and conditions are available; these separate measurements establish neither the historical result nor a commercial effect.”

UTH · Lighthouse 13.0.1 · mobile · 11 Oct 2026
RunPerformance / 100LCPCLS
1 · HTML report
99
1.18 s
0
2 · HTML report
99
1.24 s
0
3 · HTML report
100
1.07 s
0

Settings, raw values and SHA-256 hashes of all three JSON reports · Collection plan saved before the measurements.

Edikka timestamped the plan at 03:30:58.109 (UTC+2), with the first run at 03:33:59.772: 181.663 seconds later. This is not independent preregistration. The series contains three planned runs, not five; this limited repetition does not characterise all variability. We retain the original series without adding runs after observing the results.

Google counter-check: retaining the less favourable result

PageSpeed Insights, 11 October, 04:39 UTC+2: 87/100, LCP 3.7 s, CLS 0, Lighthouse 13.5.0, simulated Moto G Power, slow 4G, Chrome 153. One test executed by Google at our request; this is not independent validation of our method. CrUX reports insufficient real-user data for the tested page. Read the transcription, conditions and limitations · View the retained screenshot. Conditions differ from local runs: no pooled average or before-and-after claim.

Our conversion promise exceeded the available evidence

Claim reviewed — The earlier French edition stated “Une preuve vérifiable convertit durablement” — translated: verifiable evidence produces lasting conversion. This is a review of that actual French claim, not a fabricated client result.

Available evidence — Dated excerpt of the statement. No conversion dataset, defined event, comparison or causal design accompanied it in the page examined.

Check performed — Read the statement and look for a measurement that could test lasting conversion: conversion to what, for whom, over what duration and compared with what? The material presented does not answer these questions.

What this supports — The existence of the published promise and the lack of substantiation for that effect in the page’s evidence. The ability to inspect a source can be described as a function of the content.

What it does not prove — This does not show that evidence never influences a decision. It prevents presenting an unmeasured commercial effect as established by this dossier.

Status and decision — Not verifiable with the available evidence. Action: withdraw the outcome promise rather than attach a fictional percentage or assumed causal relationship.

Revised wording — “Relevant evidence lets a reader examine a claim and its limitations. Its effect on conversion needs separate evaluation.”

A download can establish that a deliverable is available

Claim reviewed — The promise to download the grid on our landing-page article. The French CTA reads “Télécharger la grille”.

Available evidence — Public links to the French CSV, English CSV and French JSON. Received CSV copies and their hashes are retained in the dossier.

Check performed — On 11 October 2026, HTTP requests without an account or cookies returned status 200; the files were then read and parsed. Both CSVs contain axes LP01–LP08, their criteria and expected evidence. Detailed observation and retained copies.

What this supports — Anonymous availability and the described content on that date. The central fact is that the reader can examine the grid; no commercial outcome is needed to support a download promise.

What it does not prove — This validates neither the grid’s weights, usability nor ability to predict conversion. HTTP 200 alone would not suffice: the file could be empty or unrelated. Links may change.

Status and decision — Supported within the scope “accessible deliverable and inspected content”. Action: retain the download promise without turning it into an effectiveness claim.

Revised wording — “The grid is available without an account in French and English CSV files; the files inspected on 11 October 2026 contain the eight advertised axes. Predictive effectiveness for conversion has not been established.”

Review method

Decide what you can publish

The evidence status describes the relationship between the claim and the available material. The editorial action describes what you do next. A claim that cannot be verified is not automatically false; a partly supported claim may become accurate when narrowed. The GEO benchmark observes material enabling a third-party check on 9 of 30 websites. That access and checkability finding does not classify their promises as supported: each claim must still be matched to relevant evidence. The benchmark retains its dated wave and original results.

Four evidence statuses, with no aggregate score
StatusCriterionPossible action
Supported
The available evidence supports the claim within its stated scope.
Keep it with context and limitations.
Partly supported
Evidence supports part of the claim; another part remains unsupported.
Narrow the claim or add evidence.
Not verifiable with the available evidence
Essential material is missing, inaccessible to the reviewer or does not enable the stated check.
Add evidence, remove the promise or explicitly label it as a declaration.
Contradicted
Relevant evidence contradicts the claim within the same scope.
Remove or correct it; retain the adverse result.
  1. Isolate the claim

    Copy the exact wording and its location. Separate a promised outcome, an observation and an opinion. Define what would confirm or contradict the claim.

  2. Choose relevant evidence

    Collect a report, deliverable, attributed testimonial or demonstration. Identify its source, author and relationship to the organisation. A number is needed only for a quantitative claim.

  3. Specify context and access

    Date the evidence and observation; identify the relevant page, population, period and conditions. Explain how to access the material or why access is restricted.

  4. Perform the check

    Open the evidence, inspect the deliverable or repeat the calculation. For a measurement, retain the settings and every planned run. Record failures and contradictions as well.

  5. Assess and rewrite

    Select an evidence status and explain it. Separately decide whether to keep, narrow, supplement or remove the claim. The new wording must stay within what the evidence supports.

  6. Preserve and review

    Archive the material you may retain, the check date and the decision. Name an owner and a review trigger. Check consistency across the article, FAQ and structured data.

Do not turn these categories into a score out of 100. Ticking “source present” says nothing about that source’s relevance. When reviewers disagree, retain both explanations and identify the missing evidence; averaging scores does not resolve a difference in interpretation.

Interpreting a result

Numbers, change and causality need different checks

A number needs a unit, denominator, period and method. A claim of improvement also needs a comparable baseline. A causal claim needs a design that can examine alternative explanations. These requirements do not automatically apply when checking a deliverable or a review.

Fictional example — no Edikka client data
MetricPeriod APeriod B
Sessions
2,000
4,000
Recorded enquiries
20
28
Enquiries / sessions
1.0%
0.7%

Enquiries increase by 40%: (28 − 20) / 20. The rate falls by 0.3 percentage points, or 30% in relative terms: (0.7% − 1.0%) / 1.0%. Saying “conversion increased by 40%” confuses volume and rate. These figures cannot attribute any change to adding a testimonial.

To examine an effect on conversion, define the event, lead qualification, segment, duration and decision rule in advance. Compare groups using an appropriate design, or disclose the limits of an observational comparison: acquisition, seasonality, offer, consent and tracking can change. Use the conversion funnel to locate the event being measured.

Apply the same discipline to technical results: Lighthouse is a lab measurement that varies with conditions. A score does not describe every visitor; zero WAVE errors does not certify accessibility. Keep the scope and counterexamples alongside favourable results.

What about search engines and AI?

Explicit text and consistent structured data can make evidence easier to interpret. They guarantee neither rankings, citations nor sales. Google does not require a special AI text file or markup for its AI features. Read its documentation; our Markdown files are reading formats, not evidence of search visibility.

Access and confidentiality

Share evidence without making everything public

Separate three questions: who may inspect the material, who produced it and can the check be repeated? Public data may be unverifiable; confidential material may be checked by an authorised reviewer. An independent third party may also examine evidence that is irrelevant to the claim.

Describe access without overstating the evidence
SituationWhat the reader receivesLimitation to state
Public evidence
Link to the dated report or deliverable, method and scope.
Availability observed on a given date; provenance still needs examination.
Restricted access
Redacted extract, description of hidden fields and conditions for authorised access.
A public reader cannot repeat the entire check.
Declaration only
Statement attributed to its author, with context and date.
Not corroborated by accessible evidence.
Third-party review
Report stating who examined what, with which method and conflicts of interest.
The review covers only its stated scope; independence needs documentation.

Redact personal data and commercial information that are unnecessary for the check. Do not publish client material without the relevant rights. If redaction prevents verification, disclose that limitation and narrow the conclusion instead of inventing equivalent evidence.

A SHA-256 hash lets you compare a file’s bytes with a retained hash. On its own, it proves neither authorship, creation date nor truth. Record provenance, capture date and sharing rights separately.

Documenting the review

Document a claim in your working records

Use your existing working document to record the information below for each claim. Keep the evidence, check and decision together, then name the person responsible for reviewing them. An overall credibility score cannot replace this assessment.

Fields to complete — one worksheet per claim
BlockRequired information
Claim
Exact wording, location, type and scope.
Evidence
Reference or file, provenance and author, date or period, access mode.
Check
Operation performed, date and reviewer, observation, supported conclusion and limitations.
Decision
Evidence status, editorial action, revised wording, owner and review trigger.
Only when relevant
For a number: unit, numerator, denominator and calculation. For improvement: comparable baseline. For a causal effect: design, competing explanations and uncertainty.

Edikka applies this method here to three of its own claims. Ease of use and agreement between reviewers have not been established; no independent external review has yet taken place.

To inspect the cases discussed here, read the three evidence dossiers. For assistance, send a URL and the claim to examine: the expected deliverable is a reasoned decision, not a promised uplift.

Foundations and limitations

Sources, revisions and scope of this method

Checking information and attributing sources are not Edikka inventions. Our contribution here is a documented application to three claims, a distinction between evidence status and editorial action, and explicit limitations.

To find the evidence associated with answers on this website, see our registry of documented FAQ answers.

What changed in this edition

The 11 October 2026 edition corrects the definition that required a number, withdraws the general promise of lasting conversion, distinguishes lab data from conversion and adds a new UTH observation with three reports. The internal 8 June summary does exist; our initial search missed it. It records Lighthouse 9.6.8 and an improvement-only publication rule, but does not include the raw reports. The separate Google PageSpeed counter-check returns 87/100; that less favourable result is retained with its conditions and limitations. URLs and previous anchors remain available.

These three self-audits do not establish the method’s commercial effectiveness, consistency between teams or applicability across all sectors. New evidence may change a status. To report an error, send the claim, evidence and contradicting check through the contact form.

Prompt, full source and citation

Explore this analysis.

Choose a task, review the prompt and take it to your AI.

B2B evidence: what can you really claim?Updated October 11, 2026

Read the full prompt

Your inputs stay in this browser until you paste them into your AI. No analysis runs here.

Your AI cannot read the link?

Copy the full source or download the document, then attach it to your conversation.

Download Markdown

Edikka Vision

The claim should stop where the evidence stops.

Our responsibility starts with our own claims.

When evidence is missing, we disclose it; when it supports only part of a statement, we narrow the wording. The same requirement applies to metrics and promises published by Edikka.

01 Relevance

Choose evidence that addresses the claim

A testimonial, a deliverable and a measurement establish different things. We state the conclusion each item supports.

02 Limitations

Retain the gaps as well

A missing archive, incomplete check or adverse result stays in the dossier. Hiding it would prevent readers from assessing the conclusion.

03 Reuse

Enable critical review

Dossiers and reports make our reasoning open to inspection. Publishing them establishes neither independent validation nor a commercial effect.

Key takeaway

Keep the evidence, check and limitation together, close to the claim. A precise conclusion may be less spectacular; it remains easier to examine and correct.

Article FAQ

Frequently asked questions

Additional answers to clarify the key points covered in this article.

10 selected questions View all FAQs

Does verifiable evidence guarantee higher conversion?

No. It lets a reader assess a claim and may inform a decision. An effect on conversion requires a defined event, period, comparable groups and disclosure of design limitations. The three self-audits in this article measure no commercial uplift.

How does social proof differ from verifiable evidence?

Social proof uses reviews, testimonials and client references as signals of trust. It becomes examinable when a reviewer can inspect its author, context and source. An attributed review supports an opinion; it does not establish conversion uplift on its own. Verifiability clarifies what the material actually supports.

Which evidence belongs on a B2B landing page?

Start with the objection and claim being checked. A deliverable extract can show what the service includes; a testimonial documents an opinion; a dated report documents a measurement. State the context and limitations each time. The number of evidence items is not a quality metric.

Where should evidence appear on a page?

Near the claim it supports, with a link label describing the material and its scope. Readers should be able to find the source without losing their place in the decision. Placement depends on the journey; no position guarantees a conversion uplift.

Can a testimonial provide qualitative evidence?

Yes, for an opinion attributed to its author, dated and contextualised, with provenance that can be examined. On its own, it establishes neither client representativeness, a numerical outcome nor causality. Retain publication rights and distinguish quotation, paraphrase and commentary.

Is a number or a before-and-after comparison always required?

No. An accessible deliverable or demonstration can support a qualitative claim. A number needs its definition and method; improvement also needs a comparable baseline. A before-and-after comparison alone does not prove that the intervention caused the difference.

Is a Lighthouse report or zero WAVE errors enough?

A Lighthouse report documents a run under specific conditions. Retain multiple planned runs, their settings and variability. Zero WAVE errors does not certify accessibility. Neither result demonstrates an increase in sales.

Is public evidence necessarily independent?

No. Public access describes who can inspect the material; independence describes relationships between its author, reviewer and the organisation being assessed. Repeatability is a separate dimension. Describe all three separately.

Do evidence and structured data guarantee AI citations?

No. Visible content and markup must be consistent. Google states that no special AI text file or markup is required for its AI features. A clear source remains useful to readers but guarantees neither citations nor rankings.

How do you start an evidence review?

Copy an important claim, collect the evidence, define the check and its limitations, then select separate evidence and editorial decisions. Record these details in your existing working document. Name an owner and a review trigger.

Web solutions designed to perform

Strategy. Design. Code. SEO. AI. Clearer, faster, and more compelling digital experiences.

Related articles to continue the analysis