How this site grades evidence and issues verdicts on social listening products

A social listening review methodology should make a narrow purchase decision reproducible. This one uses mandatory gates first, then scores preferences only for candidates that survive. It does not produce a universal winner.

What to take away

Review body and scope

The brandsignal.co.uk editorial desk prepared this method on 5 September 2026. No named human reviewer is assigned, so readers should treat the verdict as a desk assessment rather than an expert sign-off. No supplier paid for inclusion, no affiliate relationship exists and no other conflict was identified. Work was desk-only.

The example item is Talkwalker Social Listening for the brandsignal.co.uk editorial desk's England configuration: English-language brand research across approved social sources, analyst-reviewed coding and a monthly internal report. Talkwalker's first-party page describes visual listening, dashboards, alerts and automated sentiment. This desk check records what Talkwalker says; it has not established how those functions perform in the proposed setting.

Freeze the configuration

Record the buyer, decision, subjects, languages, source list, permitted purpose, retention, user roles, required exports, reporting audience, support window and exit artefacts. Name the entity expected to sign. Exclude functions that are interesting but unnecessary, and do not add criteria after seeing a polished demonstration.

Candidates enter only if they have a current first-party service record and can accept the same written evidence request. Reseller summaries, review stars, awards, testimonials and products lacking an inspectable record are excluded.

Grade evidence consistently

Use four grades: claimed, documented, demonstrated and buyer-tested. Record the item, date, scope and reviewer beside every grade. A sales presentation can demonstrate a task, but it cannot become a buyer test when the seller controls the data and query.

Test one evidence item at a time. A classifier check needs a frozen human-coded sample and class-specific errors. A source check needs the upstream route, allowed purpose, fields, history and removal behaviour. A support check needs a dated case with its start, severity, response and resolution evidence.

Worked example: a classifier check dated 5 September 2026, scope English-language brand posts, reviewer the editorial desk (no named individual). It is graded documented because the vendor supplied a sample but not the query.

Keep a decision log for disagreements. Before discussion, each reviewer records the criterion, evidence consulted, interpretation and proposed status. The chair documents the resolution instead of deleting the minority view.

Reopen that item if a product version, upstream source, buyer purpose, contract term or reviewer assumption changes. This stops a later team treating an old demonstration as evidence for a different configuration.

The government's AQuA Book separates verification from validation for analytical work. Its public-sector scope is explicit here; it supplies review discipline, not approval of this method or the named product.

Resolve non-compensating gates

Mark source permission, privacy roles, intellectual property, security, accessibility, classifier validity, commercial terms, exit and support as pass, fail or unknown. Do not total these results. A mandatory fail rejects the configuration, while an unknown holds the verdict.

The ICO's controller and processor contract guidance covers instructions, subprocessors, rights assistance, security, audits and end-of-contract provisions. It is under DUAA review. The buyer's named privacy and legal reviewers must assess the actual relationship and documents.

For security, use the NCSC supplier assurance questions to request scope-matched evidence and incident arrangements. Completing the questionnaire is not itself assurance.

Score preferences and state the verdict

After the gates pass, score only defined preferences such as workflow fit or training delivery. Freeze the scale, weighting and tie rule before the test. A second reviewer should be able to reproduce the result from the evidence pack.

Illustrative example: freeze a 1 to 5 scale, weight workflow fit 40 per cent, training delivery 35 per cent and support 25 per cent, and set a tie rule in favour of buyer-tested evidence. A candidate scoring 4, 3 and 4 has a weighted total of 3.65.

The configuration-specific verdict and not-reviewed list appear in What to take away.

A selection verdict or publication would require a named qualified human reviewer to resolve every mandatory gate first.

Before you act

Common questions

What is the purpose of the mandatory gates in this methodology?

The gates are non-compensating checks resolved before any preference scoring. See Resolve non-compensating gates for the full list and the rule that decides the outcome.

How does the method handle disagreements between reviewers?

It keeps a decision log. Each reviewer records the criterion, evidence consulted, interpretation and proposed status before discussion. The chair documents the resolution rather than deleting the minority view. The item is reopened when a product version, upstream source, buyer purpose, contract term or reviewer assumption changes.

Why is the verdict for Talkwalker a HOLD rather than a selection?

The verdict is HOLD because the mandatory gates remain unresolved. The specific not-reviewed items and the verdict date appear in What to take away. A selection verdict would require a named qualified human reviewer to resolve every mandatory gate first.