RESEARCH / AGENCY AI PULSE / 2026THEAGENCYAIPULSE.
218 SURVEY SUBMISSIONS
INDUSTRY RESEARCH IN CONTEXT
A pulse.
Not a census.
218 people working in and around agencies told us how AI is changing their work, their commercial models and their expectations.
The survey was open and recruited through LinkedIn and Reddit. People chose to participate. That makes the findings useful as signals of what some practitioners are seeing and anticipating, not as representative estimates for the agency industry.
Throughout this publication we separate what respondents say is happening, what they expect to happen, and what other studies report. Those distinctions matter, especially when the numbers are striking.
What respondents report
For example, whether a client has already requested an AI-related discount.
What they anticipate
For example, the pricing model they believe will dominate in two years. Not a forecast.
What other samples found
Different studies, different populations, different questions. Useful comparisons, not a combined benchmark.
AI is changing agency work.
Not everything is changing with it.
The interesting story is no longer whether agencies use AI. It is the distance between what they already do, what they trust, what they expect and what clients actually change.
Mirel Agency AI Pulse 2026, n=218. Use = regularly/most of the time; high trust = 4 or 5 on a five-point scale. Different questions.
The work is changing faster than the rules around it.
AI is already part of everyday agency practice. Every respondent reports regular or frequent use for research and insight. But adoption alone says little about whether an agency has redesigned its workflows, changed who makes decisions or found a new way to create value.
Strategy is being assisted. Judgement is not being handed over.
Nearly two-thirds use AI for strategic support; just 8% rate their trust in its strategic recommendations at four or five out of five. That is a difference in reported use and confidence, not a test of AI’s objective ability.
Agencies imagine a different future. That does not make it the present.
Seventy-two per cent would hire fewer people if they founded an agency today. Three-quarters currently name hourly/FTE retainers as their primary billing model, yet only 17% expect that model to remain primary in two years. The first is a hypothetical staffing choice; the second is a prediction about pricing, not evidence of a completed transition.
Clients may take tasks without abandoning agencies.
The GWA’s separate client-side survey finds widespread interest in bringing marketing tasks in-house, while relatively few expect agencies to play a smaller overall role. That is not a contradiction to resolve with a single number. It is a question about which work stays outside and why. [3, pp. 4–5, 22]
This is a pulse. Not a census.
The Mirel survey comprises 218 self-selected submissions recruited via LinkedIn and Reddit. Its findings describe these practitioners, not the industry as a whole. Throughout the report, OBSERVED means respondents report doing or experiencing something; EXPECTED means they anticipate it; EXTERNAL refers to a different study; and INTERPRETATION identifies a question raised by the evidence.
Everyone uses AI. That tells us surprisingly little.
Adoption is the baseline. Integration is the question.
Mirel Agency AI Pulse 2026, n=218. Research, production and strategy are regular/frequent use; agents indicate at least one in regular use. Rounded to whole percentages.
Everyone in this sample uses AI for research. That is the starting line.
↗All 218 respondents say they use AI regularly or most of the time for research and insight. Ninety-two per cent report that frequency for production. Yet those activities can still be performed as isolated tasks without changing how work moves through the agency.
Having agents is not the same as having an integrated agency.
↗Seventy-eight per cent report at least one agent or automation in regular use. That question records adoption, not whether those systems share data, are maintained reliably or form part of a connected workflow. The response option “5+ deeply integrated” was selected by 25 respondents; it combines agent count and integration depth, so it should not be read as a clean maturity measure.
The outside evidence tells a similar, but separate, story.
In the IAB’s 2025 study, 85% of its cross-industry sample reported general-purpose AI use, but 30% described AI as fully integrated into media campaigns [2, pp. 7–9]. The GWA’s agency sample reported universal generative-AI use for text and images, alongside a lower share using AI process automation [3]. Different populations and definitions prevent a like-for-like adoption comparison.
The next dividing line may be operational, not technological.
What matters next is less whether an agency has an AI tool and more whether it has a dependable way of using it across real work. That is a proposition for future measurement, not a result established by the Pulse.
Hourly billing isn't dead.
Agencies are already imagining its funeral.
The expectations are dramatic. The evidence of completed change is much thinner.
Mirel Agency AI Pulse 2026, n=218. Primary billing model selected for today and for two years ahead. Rounded to whole percentages. Future responses are expectations, not observed changes.
have experienced AI-related client discount requests
Clients have started asking for their AI discount.
Asking for a discount is not the same as getting one. In Productive's separate agency sample, 27% reported discount requests, while 13% said they had actually lowered prices to remain competitive.
Different samples and questions. These percentages do not establish a trend between studies.
The expected pricing revolution has not happened yet.
Today, 76% name hourly or FTE retainers as their primary billing model. Only 17% expect to name them in two years. Respondents instead expect fixed/project pricing (33%) and value/performance pricing (34%) to become much more common as their main model. These are answers to two separate questions about primary billing, not a forecast of revenue shares.
A discount request is not a discount granted.
↗In the Pulse, 53% report that a client has asked for a lower price because of AI. That says nothing about whether the agency agreed. In Productive’s separate 2025 customer survey, 27% reported discount requests and 13% said they had lowered prices [1, pp. 17, 20–21]. The difference between those samples cannot be interpreted as a trend.
Efficiency can improve the margin without changing the invoice.
Productive found that 65% of its respondents reported at least one positive revenue or profitability effect from AI [1, p. 15]. That does not mean new AI services are generating revenue: only 13% reported meaningful new revenue streams [1, p. 25]. Delivery efficiency, price concessions, pricing model and new revenue are four different questions.
Agencies may be imagining a harder transition than they realise.
Promethean Research reports that value-based pricing adoption fell from 31% of agencies in 2024 to 23% in 2025 [4]. Its definition and sample are separate from the Pulse. The juxtaposition raises a practical question: if agencies want to sell outcomes, what will they measure, contract for and take responsibility for?
The agency they would build today has fewer people.
An imagined first-year team is not the same as today’s payroll.
Mirel Agency AI Pulse 2026, n=218. Hypothetical first-year hiring if founding an agency today; not actual headcount reductions. Percentages rounded.
The most revealing staffing question is hypothetical.
Asked how many people they would hire in their first year if founding an agency again today, 44% chose 20–40% fewer and 28% chose at least 50% fewer. Another 28% chose about the same. This is a redesign thought experiment, not evidence of jobs already lost.
A smaller imagined agency is not a redundancy programme.
Productive reported that 64% of its customer-agency respondents had made no AI-related redundancies and had no plans to do so [1]. In the GWA’s agency sample, 13.8% reported plans for AI-related savings or redundancies [3]. Neither result measures what Pulse respondents would do if starting from scratch.
The first change may be fewer new hires.
↗Agencies could reduce junior hiring, redesign roles, shift work between disciplines or let attrition change team shape. None of these pathways can be inferred from the hypothetical hiring answer alone. A future Pulse should measure new hires, attrition and redundancies separately.
AI is in the strategy room. It isn't in the driver's seat yet.
AI is being used to support strategic work far more often than its recommendations are highly trusted.
Mirel Agency AI Pulse 2026, n=218. Share rating trust 4 or 5 on a five-point scale, rounded from 69.7% and 8.3%. Separate self-reported questions.
AI is helping with strategy. Trust is another matter.
Sixty-four per cent report regular or frequent AI use for strategic support. Only 8% give AI strategic recommendations a high trust rating (4 or 5 on a five-point scale). For production work, the equivalent high-trust share is 70%. These are respondents’ confidence ratings, not objective assessments of output quality.
The disclosure question is more complicated than a yes or no.
Forty-eight per cent say they have presented AI-assisted strategy to a client without telling them. The survey does not ask whether disclosure was required, expected or material to the work. It would be irresponsible to treat the answer as evidence of deception. It does suggest that client expectations and disclosure practices deserve closer examination.
Who makes the final call may matter more than who drafts the recommendation.
↗Research, synthesis and ideation can all benefit from AI without transferring accountability. The IAB documents substantial AI use in structured media tasks, while the GWA client findings show why the perceived value of agency expertise remains a live question [2, 3]. The interesting next measure is decision authority: when does AI propose, when does a human approve, and who owns the result?
The creative versus performance divide might be the wrong one.
Agency labels may be a poor shortcut for understanding what AI can actually automate.
Mirel Agency AI Pulse 2026. Creative agency n=49; performance/paid media n=43. Exploratory subgroup comparison; small, self-selected samples. Rounded percentages.
The expected gap between creative and performance agencies is not clean.
Among creative/brand respondents (n=49), 71% report agents in regular use; among performance/paid-media respondents (n=43), 77% do. Client discount requests are reported by 47% and 58% respectively. These are exploratory comparisons from small, self-selected groups, not population differences.
The type of work may matter more than the agency label.
Performance workflows often contain repeated inputs and measurable outputs, making some forms of automation easier to define. But creative agencies also have repeatable processes, and performance agencies make consequential judgement calls. The Pulse does not measure task-level repeatability, so this remains a hypothesis worth testing rather than a conclusion.
Integration still has practical obstacles.
In the IAB’s 2025 research, respondents identified setup complexity and security, gaps in AI knowledge, and fragmented technology as major barriers to scaling AI [2, pp. 15–16]. Those obstacles may cut across agency categories more sharply than any creative-versus-performance divide.
AI can make agency work faster. Who gets the value?
AI creates efficiencies. The distribution of their value is still being negotiated.
Conceptual framework illustrating possible commercial outcomes, not a statistical model. Survey and external-study results discussed in the text.
A faster deliverable does not tell us who wins.
Fifty-three per cent of Pulse respondents have received AI-related discount requests. Forty-five per cent report clients bringing some work in-house because of AI. Neither answer measures the revenue lost, margins retained or value passed to clients. The commercial outcome depends on what work moves, what remains and what agencies can charge for it.
Clients may take work in-house without wanting fewer agencies.
In the GWA’s separate survey of 170 company-side respondents, 80% said they wanted to bring more marketing services in-house using AI, yet only 8.2% expected agencies to play a smaller overall role [3, pp. 4–5, 22]. These are intentions and expectations, not observed budget transfers. They suggest the unit of competition might be a task rather than an entire relationship.
Agencies can capture value in more than one way.
Productive’s research suggests some agencies are already seeing revenue or profitability benefits, while relatively few report meaningful new AI-generated revenue streams [1, pp. 15, 25]. The IAB also records agency concern about client in-housing and brand concern about partner transparency [2, p. 21]. Keeping margin, winning new work and selling a new service are different outcomes.
The value question is still open.
Clients may demand savings, platforms may absorb tasks, and agencies may retain margin or sell expertise that becomes more valuable when execution is cheaper. The Pulse cannot rank those outcomes. Future research should track prices actually paid, agency margins, client spend and work transferred.
“we should be interested in where ai is actually replacing work rather than where agencies say theyre using it.”Reddit commenter
“At my old agency we used it mostly for first drafts of ad copy and email subject lines. Not replacing real writers but saving them from staring at blank page.”Reddit commenter
“Compressing delivery timelines and exposing inefficiencies in our hourly retainer model.”Mirel Agency AI Pulse 2026, open-ended response
“Allowing small integrated teams to deliver the output of agencies twice our headcount.”Mirel Agency AI Pulse 2026, open-ended response
“Raising the quality bar for concepting while demanding tighter human editorial filters.”Mirel Agency AI Pulse 2026, open-ended response
“If you’re not using AI in your agency you’re wasting your time and client money. As an example, I have a PPC agent monitoring client campaigns...”Reddit commenter
“Helping us draft rapid press angles while keeping media relations deeply human.”Mirel Agency AI Pulse 2026, open-ended response
Selected verbatim comments are illustrative. They are not treated as a statistically coded qualitative dataset.
Questions worth coming back to.
The most useful follow-up is not another adoption statistic. It is evidence of what actually changes.
The Pulse is most useful as a set of questions to revisit. A future edition could track whether the current expectations become observable changes.
The GWA’s parallel client survey adds a further variable for follow-up: whether clients take more individual tasks in-house while maintaining or increasing their overall reliance on agencies for other work. Tracking task transfer, total agency spend and perceived strategic value separately would help resolve this apparent contradiction.
| Signal | 2026 baseline | What to measure next |
|---|---|---|
| Pricing | 2026: 76% hourly/FTE primary today; 17% expect it in two years. | Actual primary model, revenue share and realised prices in 2027–28. |
| Staffing | 2026: 72% would hire fewer if founding again. | Junior hiring, total hires, attrition and redundancies separately. |
| Integration | 2026: 78% report agents in regular use. | Connected workflows, uptime, maintenance and degree of autonomy. |
| Strategy | 2026: 64% use AI for strategy; 8% highly trust recommendations. | Decision authority, human review, disclosure expectations and trust. |
| Client pressure | 2026: 53% report discount requests; 45% report in-housing. | Discounts granted, revenue affected and tasks actually transferred. |
| Value capture | 2026: survey does not measure realised margins. | Gross margins, client spend, revenue per employee and new-service revenue. |
What these numbers can and cannot tell us.
The Mirel Agency AI Pulse 2026 collected 218 submissions through an open online survey promoted on LinkedIn and Reddit. Participation was voluntary, making this a self-selected, non-probability sample. The results are indicative of the respondents’ views and experiences rather than representative of the agency industry.
The respondent mix is weighted towards independent agencies (182/218), smaller organisations (165/218 report 2–50 employees) and UK/Ireland, Western Europe and North America (212/218). Those characteristics matter when interpreting the results. The findings describe reported experiences, behaviours and expectations. They do not establish causal effects or reliably predict future outcomes.
Productive’s Agencies in the AI Era: Between Hype and Reality is based on 181 agencies and consultancies using Productive, surveyed online from 9 to 16 September 2025. Most respondents were in Europe and 63% held executive leadership roles. Productive describes its findings as indicative of its customer base, not the wider professional-services industry [1, p. 32].
The analysis uses whole-sample distributions as its quantitative foundation. Agency-type comparisons are exploratory. The open-ended survey answers are not statistically coded; selected verbatim comments illustrate individual views only. All 218 submitted responses are retained.
Existing industry research is used to corroborate, complicate and extend the Pulse findings, not to convert a self-selected sample into a representative one. Different studies use different samples, geographies, definitions and periods. Their percentages should not be treated as directly comparable without examining the original research instruments.
The GWA KI-Whitepaper 2026 draws on two independent descriptive online surveys: 78 GWA-member agency leaders (12 February–6 March 2026) and 170 company decision-makers (1–9 April 2026). Neither sample is claimed to be representative, and the two groups cannot be linked respondent by respondent [3, p. 7]. Its percentages are presented here with their own denominators and question contexts; they are not pooled with the Pulse data.
The wider picture.
External research is cited for context, not presented as a pooled sample or endorsement.
[1] Productive. Agencies in the AI Era: Between Hype and Reality (2025). Survey of 181 Productive customer agencies and consultancies, 9–16 September 2025. Especially pp. 6, 11, 15, 17, 20–21, 25 and 32. https://productive.io/agencies-in-the-ai-era-between-hype-and-reality/
[2] Interactive Advertising Bureau (IAB), BWG Global and Transparent Partners. State of Data 2025: The Now, the Near, and the Next Evolution of AI for Media Campaigns (March 2025), especially pp. 7–10, 15–16, 19, 21, 24–25. https://www.iab.com/news/iab-state-of-data-report-2025/
[3] GWA Tech & Innovation Forum. DER NEUE WERT: Agenturen im Spannungsfeld von KI-Reife und Kundenerwartung. GWA KI-Whitepaper 2026, Ausgabe 3. Two descriptive surveys: agencies n=78 and client-side organisations n=170.
[4] Promethean Research. Value-Based Pricing Was Supposed to Save Us. https://prometheanresearch.com/value-based-pricing-was-supposed-to-save-us/
Mirel Thinking
Further reading.
- The economics of strategic work in creative agencies
The commercial questions behind strategic work.
- Does AI replace strategists? It replaces some strategic labour.
How AI changes strategic labour and human judgement.
- Can AI do strategy?
A closer look at what AI contributes to strategy.
Publication record: DOI: 10.5281/zenodo.23689848