Branding agency claims: how to read UK evidence

Case	What the record can support	What it cannot establish
Government plans	Published ambitions and possible routes to monitor	Funded briefs or agency revenue
Business registers	Counts of enterprises and local units by industry and region	A count of branding agencies, buyers or live demand
SIC revisions	Classification context for a dated market table	A retrospective England branding-agency series
Evidence classes	The scope and limits of a particular claim	A single, blended measure of future outcomes
AI surveys and plans	Reported business use and published recommendations	England agency adoption or proven commercial improvement
Copyright policy	Evidence, options and planned follow-up	New rights law by publication alone
Tool descriptions	Possible functions to assess in a workflow	Accuracy, authorisation or suitability of a product
Occupation projections	National trends and skills questions	Staffing levels at England branding agencies
Market and client signals	Records to qualify and monitor	A complete pipeline from one signal
Scenarios	Conditional choices with triggers	Predictions about what will happen
Publication boundary	The dates and limits of the evidence reviewed	A forecast of demand, performance, prices or adoption

Treat public ambition as context, not booked agency work

When writing or reading a directory profile or review of a branding agency, treat public ambition as context, not evidence of that agency's booked work. The UK government's Creative Industries Sector Plan, published on 23 June 2025 and updated on 5 August 2025, sets ambitions around investment, technology, resilience and skills across the UK creative industries. The plan alone establishes neither a funded brief nor expected agency revenue.

Before treating a policy initiative as evidence of booked work, require a named fund, published buyer notice, eligibility rule, named budget authority, timetable and signed opportunity relevant to the agency's service and English operating area. If any element is missing, retain the plan as context rather than pipeline. This is a decision threshold, not a forecast.

Read supplier counts at their actual unit

ONS business counts describe registered enterprises and local units, not live briefs, purchasers or completed projects. Industry definition and geography also matter: branding activity can sit inside mixed-service firms or in-house teams. Broad business categories therefore do not provide a clean count of branding agencies or their buyers, or verify the completeness or quality of a directory listing or review.

The Office for National Statistics released UK Business: activity, size and location 2025 on 24 September 2025. Its enterprise and local-unit tables use an Inter-Departmental Business Register snapshot from 14 March 2025 and can be analysed by industry and region. They support those described counts, not a measure of demand.

A head-office location does not show where every client decision or delivery team sits. For directory editors and review readers, ICAEW's UK industry profile offers category context: it describes Creative and brand agencies (focused on developing advertising ideas and content, and/or brand identity) as a segment of the UK advertising and marketing industry, not a separate England count.

SIC 2026 Implementation

14 March 2025
IDBR snapshot taken
24 September 2025
ONS business release
April 2026
SIC 2026 revision finalised
2031
Earliest National Accounts use

Keep classification changes visible

Name the SIC version on every 2027 market table. ONS UK SIC 2026 guidance says the revision was finalised in April 2026, but it cannot enter outputs immediately. Implementation moves through business registers, survey sampling and National Accounts; 2031 is the earliest anticipated use in National Accounts.

Do not join SIC 2007 history to a SIC 2026 extract without the relevant correspondence, metadata and a recorded assessment of the break. A revised classification may improve later measurement, but it creates no retrospective England branding-agency series. No defensible numerical forecast follows from these records; see the branding agencies outlook.

Keep evidence classes distinct

Observed data, announcements, forecasts and scenarios support different claims. Label each future-facing claim before relying on it, and retain the source details that show what can be inferred and what still needs checking. Combining unlike evidence can turn a qualified record into an error chain.

Four Evidence Classes

Observed data

What it is: Measurement already made
Key check: Publisher and limits
Common error: Not a buyer

Dated announcement

What it is: Published policy or rule
Key check: In force or proposal
Common error: Not delivery

Attributed forecast

What it is: Organisation's projection
Key check: Base period and model
Common error: Wrong geography

Editorial scenario

What it is: Conditional planning device
Key check: Triggers and review date
Common error: Not a prediction

Evidence label	What it can establish	What to retain or check
Observed data	A measurement already made	Publisher, population, geography, collection period, release date, classification, known error or coverage limit
Dated announcement or requirement	Published policy, rule, standard or intended action	Whether it is in force, guidance, consultation, proposal or independent recommendation; publication is not delivery
Attributed forecast	The producing organisation's projection	Base period, horizon, model, assumptions and uncertainty; do not transfer geography or population
Editorial scenario	A conditional planning device	Assumptions, observable triggers, owner, response, stop condition and review date; it is not a prediction

For observed data, record its publisher, population, geography, collection period, release date, classification and known error or coverage limit. These details define what the evidence describes and help prevent a broad count from being mistaken for an estimate of a narrower agency market.

A dated announcement or requirement describes published policy, a rule, a standard or intended action. Check whether it is in force, guidance, consultation, proposal or an independent recommendation. An announcement records publication or intent; it does not demonstrate delivery or implementation.

An attributed forecast belongs to the organisation that produced it. Retain its base period, horizon, model, assumptions and uncertainty. Do not transfer a UK sector projection into England, or treat a projection for a broad occupation group as a direct measure of branding agencies.

An editorial scenario is a conditional planning device, not a prediction. State its assumptions and observable triggers, name an owner and response, define a stop condition, and set a review date. These distinctions help avoid measurement mistakes: policy ambition does not prove investment, investment does not establish agency demand, tool capability does not show adoption, and adoption does not demonstrate a better client outcome.

Read AI findings at their stated scope

UK-wide survey results cannot be relabelled as an England branding-agency adoption rate. The ONS material covers businesses across broad industries; it does not isolate branding agencies, establish that reported use improved commercial outcomes, or provide an agency-specific baseline.

The ONS article Artificial intelligence in UK businesses: 2023 to 2026 was released on 20 July 2026. Its latest analysis uses the Business Insights and Conditions Survey, Wave 159, fielded from 5 to 28 June 2026. It covers UK businesses across broad industries and discusses reported uses, barriers and effects.

That record may support a UK business or broad-industry baseline if its sampling and response limits are carried forward. It reports respondents' answers, not a technical audit of their systems or proof that an AI use caused commercial improvement. Do not present it as evidence about an England agency population.

The AI Adoption Plan for Creative Industries was published on 8 June 2026 as an independent report by the sector's AI Champion. Its proposals cover responsible adoption, guidance, skills and workforce transition across the UK creative industries. They are recommendations, not enacted requirements or evidence that branding agencies have implemented them.

Separate copyright work from current rights

A policy workstream is not a new rights rule. Treating it as enacted law can turn a watch item into legal advice. Keep the current legal and contract position distinct from consultations, government responses, proposed legislation and any commencement date.

The government's Report on Copyright and Artificial Intelligence was presented in March 2026 after the 2024 to 2025 consultation. It reports evidence, policy options and planned follow-up, rather than making new law by itself. Its subjects include input and output transparency, licensing, computer-generated works and digital replicas.

For a 2027 brief, record the current UK rules and contract position, then track consultations, responses, legislation and commencement separately. A named UK IP lawyer should approve any statement that turns policy material into operational rights advice. Recheck the position on publication day rather than copying an earlier compliance statement.

Test each proposed tool workflow

A described capability does not establish that a particular product is accurate, authorised or suitable. A tool may generate text or images, retrieve or summarise information, and suggest drafts, but each proposed workflow still needs its own checks and an accountable decision owner.

The Government Digital Service's AI Playbook, published on 10 February 2025 for government and public-sector organisations, describes systems that can generate text and images, retrieve or summarise information, and suggest drafts. Its description of possible functions does not verify an agency's product, process or outputs.

A bounded scenario might use a controlled system to organise an approved research corpus, propose internal copy variations or create a clearly marked concept for human assessment. Compare the workflow with a manual baseline, recording inputs, model and version, permissions, output reviewer, rejected material, corrections, time spent and exit route.

Put each check behind a named gate:

Human gates

IP review checks training, prompt and output material, licences, resemblance and contract restrictions.
Privacy review checks whether personal information enters the workflow and on what role, purpose and controls.
Advertising review tests every objective claim and the impression created by the finished communication.
Accessibility review examines the actual asset and user route, not whether a tool offered an accessibility feature.
Security review covers supplier access, retention, model or prompt attacks, logging, incidents and deletion.
Human review decides whether the result meets the brief and may be used at all.

Copyright and licensing questions refer to the Copyright, Designs and Patents Act 1988 and UK Intellectual Property Office guidance. Personal data falls under the UK GDPR, the Data Protection Act 2018 and the ICO. Advertising claims are assessed against the CAP Code, the UK Code of Non-broadcast Advertising and Direct & Promotional Marketing, enforced by the Advertising Standards Authority.

Accessibility review draws on the W3C Web Content Accessibility Guidelines and Equality Act 2010, plus the Public Sector Bodies (Websites and Mobile Applications) Accessibility Regulations 2018 for public-sector clients. Technical security questions can use NCSC secure AI guidance. These references do not replace review of the actual asset, workflow, supplier or contract.

The ICO's AI and data-protection guidance covers accountability, transparency, lawfulness, fairness, security and rights. Its page says the guidance is under review following the Data (Use and Access) Act. A 2027 workflow therefore needs a publication-day recheck, not a statement copied from an earlier date.

The NCSC's international secure AI development guidance is organised around secure design, development, deployment, operation and maintenance. Version 1.0 was reviewed on 27 November 2023. It supplies security questions, not certification of an agency or a guarantee about a hosted tool.

Use occupation projections to frame questions

A national occupational projection is not a forecast of staffing at England branding agencies. Using it as a direct payroll plan changes both the population and the unit. Treat the projection as context for skills questions, not as an agency headcount or employment decision.

Skills England's 2026 creative-industries assessment, updated on 4 August 2026, projects demand through 2035 for 30 UK creative-industry priority occupations, including graphic and multimedia designers. Its occupational codes are approximations, and its projections indicate trends and orders of magnitude rather than precise employment levels.

The technical annex says the creative-industries projection uses assumptions about investment, gross value added, productivity and occupational prevalence. Neither the assessment nor the annex measures future staffing at England branding agencies. Validate workforce choices against signed work, capability evidence and employee consultation instead.

Use the projection to frame questions: which tasks need creative judgement, research or commercial skill; who can challenge an automated output; and where do accessibility, rights and security competence sit? Check the answers against confirmed work and assessed capability, rather than inferring an agency staffing requirement from the national projection.

Build qualified demand from several records

A single market, client, technology or skills signal cannot stand in for a complete pipeline. Give each measure an owner, retain its underlying record, and state what it can and cannot show before using it to change plans.

Market evidence:monitor official releases, SIC version, geography, service definition and whether a count describes enterprises, jobs, spend or output in the 2027 market evidence. Stop comparisons when the unit changes.
Client work:track qualified enquiries, brief types, procurement notices, win and loss reasons, contracted scope and cancellations. Keep personal data out of planning views unless necessary and authorised. More enquiries are not realised demand until buyers qualify and contract.
Technology:log approved uses, manual baseline, failure categories, review time, rights questions, incidents and client restrictions. Faster generation is not a productivity gain if correction or assurance work has simply moved elsewhere.
Skills:compare confirmed work with assessed capability, availability, training completion and supervised performance. Do not use a national occupational forecast to make an individual employment decision.

Make uncertain futures conditional

A scenario supports a conditional choice, not a claim about what will happen. Give it an observable trigger, an owner, a response, a stop condition and a review date. This lets uncertainty guide a bounded decision without presenting a planning assumption as a forecast.

Controlled augmentation:consider a limited pilot if clients permit a defined AI use, licences and data terms are clear, and synthetic tests meet agreed acceptance criteria. Stop for unexplained resemblance, personal-data leakage, inaccessible output, unsupported claims or an unrecoverable security event; see controlled augmentation.
Evidence-heavy procurement:respond when buyers require fuller provenance, accessibility, security, subcontractor or AI-involvement records. The commercial owner should maintain a current evidence pack and refuse unsupported claims. Restart a held bid only when the missing record has a named issuer and applicable scope.
Delayed adoption:keep the manual route when client prohibitions, unclear rights, weak controls or correction effort outweigh tested benefit. The operations owner records the reason and schedules a new review after a source, contract or product change. Delay is a valid decision, not a failure to follow a presumed trend.

Fix the evidence boundary to a date

An undated outlook can conceal changes to sources, classifications and rules. Set the publication boundary, then recheck relevant records on publication day. Do not give an evidence-based appearance to a claim whose source date, geography or measurement unit has changed.

The public records considered here were available by 5 September 2026. Recheck ONS releases, classification implementation, official policy status, regulator guidance and each cited record before publication. Date scenario decisions and attribute forecasts to their producers, preserving their geography and method rather than transferring them to a different population.

This is an evidence-reading guide for claims that may appear in branding-agency directory listings or reviews, not a directory, rating or review of named agencies. No agency survey, customer interview, private tender analysis or forecasting model was conducted. The evidence supports no adoption rate, market size, price trajectory or probability. It does not predict England branding-agency demand or performance; approve ninety-day actions only against observed buyer evidence and separate legal, rights, accessibility, privacy, security, workforce and commercial gates.

Use these checks before publishing or relying on a directory profile or review of a branding agency:

Label each future-facing claim as observed, announced, forecast or scenario.
Record publisher, population, geography, period and limits for observed data.
Check whether policy is in force, guidance, consultation or proposal.
Name the SIC version on every market table.
Do not book revenue from a policy document alone.
Put AI outputs behind IP, privacy, advertising, accessibility, security and human gates.

Keep copyright ownership of source, training, prompt and output material within the rights review. Record data sharing and any relevant contract restrictions before a workflow is approved. The evidence pack should also distinguish the CAP Code, Web Content Accessibility Guidelines, Equality Act 2010 and Accessibility Regulations 2018 where those references apply.