In short
- The model is no longer the difference. The context it receives at the moment of a decision is.
- Most marketing teams are still at the stage of first prompt templates; adoption of AI runs far ahead of the checks on its output.
- Marketing operations should decide which context matters most, write it down once per brand, hand each task only what it needs, and keep a record of what the model was given.
1AI is a commodity. Context is not.
Every marketing team has access to the same models. They write fluent text in many languages, summarise a hundred articles in a minute and answer questions about almost anything. What they do not know is your brand: what you may and may not say, who is speaking, which product facts are true this quarter, what the press has already written and what your competitors claim.
The authors of the State of Martech 2026 report draw the consequence from an old saying, that you don’t have an AI strategy without a data strategy:
We’d expand that: You don’t have an AI strategy if you don’t have a context strategy. AI is a commodity. Context is differentiation.
Elsewhere the report makes the same point about speed: language models became a commodity faster than the context companies hold about themselves. The advantage, the authors argue, lies less in the choice of model than in how well a company’s context is connected and delivered.(State of Martech 2026, p. 36)
2What context engineering means
The term moved from AI engineering into marketing within a year. In June 2025 Philipp Schmid described it as the discipline of building systems that hand a model the information and tools it needs for a task, in a usable format and at the moment it needs them. State of Martech 2026 brings the idea to the place where marketing feels it, the moment a decision is made:(Philipp Schmid, 2025)
Context engineering is the work of making context usable at decision time. It means assembling the right data, policies, tools, permissions, and instructions so that an agent or application can act intelligently in a specific use case. Not just more data. Better-prepared context.
The report places this work between three kinds of context. Company context is what the company knows about itself: goals, strategy, brand, processes, governance. Customer context is the customer’s situation, intent and history. Systems context is what the company’s systems can actually access, connect and deliver. Context engineering covers the overlap of systems with the other two: the place where what a company knows becomes readable for machines and what it understands about customers can be acted on.(State of Martech 2026, pp. 59–60)
Where context engineering sits
- Company context
- Goals, strategies, brand, processes, capabilities, governance and priorities.
- Customer context
- The customer’s situation, goals, intent, preferences, history and moments that matter.
- Systems context
- What the company’s systems can actually access, connect and deliver.
Context engineering covers the overlap of systems with the other two. Where all three meet, the report speaks of golden context.
Company, customer and systems context; context engineering is the overlap of systems with the other two. Drawn after State of Martech 2026.
One warning in the report matters for every team that starts pasting documents into a chat window: more input does not automatically mean better output. It can be worse, and more expensive. Context engineering means selecting what a task needs, not piling on everything there is.(State of Martech 2026, p. 61)
3Context only counts at decision time
Context has no value in storage. It only creates value when it arrives at the moment of decision — when a system, an agent, or a person has to choose what to do next for a specific customer in a specific moment.
Most brands already have the material. The brand book is a PDF, the product facts are in a spreadsheet, the messaging lives in last quarter’s campaign brief, the press coverage sits in a monitoring tool. The report’s image for this is blunt: an insight that nobody uses is not an asset but stock going off on the shelf.(State of Martech 2026, p. 66)
Earlier software solved this with rules hard-coded into every system. Agents change where the work goes, not how much of it there is: the report notes that they do not remove that complexity but move it somewhere else. The rules move out of the application and into context that has to be structured well enough for a model to read it at runtime.(State of Martech 2026, p. 67)
One decision, the context it needs
- Brand voice and compliance rulesBrand profile
- Who speaks, and what to avoidAuthor persona
- Who it is written forAudiences
- Current specifications and pricesProducts
- Key messagesCampaign brief
- What the press already wroteCoverage
Decision time
The model writes the press release
A press release for a product launch: what has to reach the model at the moment it writes. Schematic.
4Context moves at different speeds
Not all context changes at the same pace. Borrowing Stewart Brand’s pace layers, the report stacks marketing context from slow to fast: market, company, relationship, journey, session and moment. The slow layers are stable and widely available; the fast ones decay within hours or minutes and are, the authors argue, where differentiation now lies.
Context stack pace layers
- Moment secondsreal-time state, current queryYour website, chat and analytics
- Session minutes to hoursactions, conversation, content consumptionYour website, chat and analytics
- Journey days to weeksbuying stage, signals, engagement patternYour CRM, CDP or marketing automation
- Relationship months to quartersaccount history, preferences, lifetime valueYour CRM or CDP
- Company quarters to yearsbrand, strategy, capabilities, governanceBrand profile, speakers, audiences, products, knowledge bases
- Market yearsindustry structure, macro trends, regulationsCompetitors, topics, coverage and AI answers, collected daily to weekly
Layers and time scales after State of Martech 2026. The right column shows where communication context sits and which layers stay in other systems.
For communication and PR the practical point is that the two slow layers are not slow anymore. Competitors launch, the press writes, AI assistants change how they describe a brand. Company context, the brand’s own rules and facts, has to be current at every draft. Both layers have to be kept up to date continuously, not once a year in a brand book.
5Where marketing teams are today
Adoption is not the problem. In the State of Martech survey of 208 marketing and marketing operations leaders, AI use rose in every category of the martech landscape between 2024 and 2026. The authors read their sample as the leading edge rather than the median.(State of Martech 2026, pp. 29–30)
AI adoption by martech category
Share of respondents using AI in the category, 2024 and 2026; n = 208 in 2026
- 2024
- 2026
- Content & Experience
- 2024: 79 %, 2026: 89 %
- Data
- 2024: 61 %, 2026: 75 %
- Management
- 2024: 58 %, 2026: 72 %
- Advertising & Promotions
- 2024: 30 %, 2026: 50 %
- Social & Relationships
- 2024: 33 %, 2026: 49 %
- Commerce & Sales
- 2024: 28 %, 2026: 49 %
Content is where AI arrived first and goes furthest.
The checks did not keep pace. The report calls it the governance gap: in Content & Experience, AI copy production stands at 91 % adoption, while content authenticity and AI detection stand at 37 %. 103 of 163 respondents do nothing to verify the AI content they produce at scale.(State of Martech 2026, p. 34)
The governance gap in content
Adoption among respondents, Content & Experience
91 %
AI copy production
37 %
Content authenticity and AI detection
A survey in Germany, Austria and Switzerland shows the same picture from the context side. Asked about their maturity in context engineering, three quarters of marketing decision-makers place themselves at level 1 or 2: AI used now and then, or first guidelines and prompt templates. Three percent have context engineering built into systems and workflows.(Marketing Tech Monitor 2026, fig. 5)
Maturity in context engineering
Share of marketing decision-makers, Germany, Austria, Switzerland; n = 414
- 1Ad hoc: Generative AI used now and then, no systematic use of context, 20 %
- 2First rules: Importance of context recognised; first guidelines or prompt templates, 55 %
- 3Structured: Context engineering planned and part of processes, 22 %
- 4Integrated: Part of systems and workflows; data sources used automatically, 3 %
- 5Orchestrated: Fully strategic, scalable and optimised, 0 %
Level 4 is where context reaches every operation without being pasted in by hand.
6Beyond marketing: what CEOs see
The question is not a marketing question alone. EY-Parthenon’s CEO Outlook of October 2026 asked 1,200 CEOs of large companies in 21 countries what drove their productivity gains over the past twelve months. AI-enabled tools came first, but business process redesign came close behind.
What contributed most to productivity gains
Share of CEOs, global, past 12 months; n = 1,200
- AI-enabled tools50 %
- Business process redesign46 %
- Cost and asset optimization41 %
The report’s reading: the new technologies pay off most where the way of working is redesigned around them, and least where they are added on top of unchanged processes.(EY-Parthenon CEO Outlook, October 2026, chapter 1)
Gains do not turn into value by themselves. 48 % of the CEOs say productivity gains are reinvested in growth, innovation or transformation; 23 % say they are absorbed by complexity, regulation, risk or new work demands. Only 16 % have comprehensive, real-time visibility into AI costs and high confidence in assessing the return.(EY-Parthenon CEO Outlook, October 2026, chapters 1 and 2)
Where productivity gains go
Share of CEOs, global; n = 1,200
48 %
Reinvested in growth, innovation or transformation
23 %
Absorbed by complexity, regulation, risk or new work
The most striking line of the report is about work itself. It expects roles to change before headcount does, and describes the change in terms of context:(EY-Parthenon CEO Outlook, October 2026, chapter 2)
As a result, roles will increasingly be defined not by the tasks associated with them, but by the knowledge requirements and constraints of their associated contexts.
7Where marketing operations teams should focus
State of Martech 2026 counts aligning company context, customer context and systems among the central jobs of marketing operations and martech leadership over the next 12 to 18 months. It sees the marketing ops team becoming the context engineer: the people who decide which data, content, tools and instructions reach each AI agent, and when. In practice that means six things, in this order.(State of Martech 2026, pp. 3, 62)
- 1
Decide which context matters most, and connect it first
Not every document. For communication that is usually the brand’s rules and voice, who speaks, the product facts, the approved messages and what the market is saying. The report’s advice is to settle on the context that matters most and to connect that first.
- 2
Write the slow layers down once per brand
Brand voice, compliance rules, speaker personas and product facts belong in one maintained place per brand or client, readable by machines, not pasted into each prompt by whoever writes it.
- 3
Hand each task only what it needs
A social post needs different context than a quarterly report. Selecting and sizing the context per task is part of the job: selection, not volume.
- 4
Check the output and keep a record
Close the governance gap: check AI drafts against facts and briefs before review, and record which instructions and which model made each analytical decision, so that it can be explained later.
- 5
Redesign the workflow, do not bolt AI on
Remove hand-offs that only existed to copy information from one tool into the next. Keep people where judgement matters: approval, legal review, client sign-off.
- 6
Measure outcomes, not adoption
EY’s advice to CEOs applies to marketing operations too: tie the time saved to output, margin, working capital, customer experience or resilience. The number of prompts run is not a result.
The problem isn’t building software. It’s making sure that software knows enough, is allowed enough, and is constrained enough to make good decisions.
8Why Acontema is built the way it is
Acontema was built on the assumption these studies describe: that the model is replaceable and the context is not. That is why the platform starts with a project per brand or client, not with a text box, and why every AI step reads from it.
- Decide which context matters, and keep it in one place
- One project per brand or client: brand profile, author persona, audiences, competitors, knowledge bases, products and coverage. Every operation reads from it.
- Make company knowledge machine-readable
- A brand profile is generated from the website in about a minute and then edited by the team; style is learnt from example posts. Documents, scanned PDFs, links and YouTube subtitles become searchable knowledge, analysed for topics and entities.
- Select, do not pile on
- For each task the system picks the pieces that matter, sizes how much of each source fits, and passes them in a fixed structure.
- Constraints the model must respect
- Compliance rules in the brand profile (disclaimers, prohibited topics, sensitive terms) and an “avoid” list per speaker. The AI does not invent experiences beyond those in the persona.
- Context that does not spoil
- Coverage is collected daily, every three days or weekly; competitors are re-analysed on a schedule. Watching agents check collection and analysis coverage and flag what needs attention.
- Check and record
- Drafts are checked against campaign, audience and facts before review. Prompts for analytical decisions are versioned, and each decision is stored with the prompt version and model that made it. Approvals run per language version.
Sources
- State of Martech 2026Scott Brinker and Frans Riemersma (chiefmartec, MartechTribe), May 2026. Survey of 208 marketing and marketing operations leaders on about 70 AI use cases, February 2026.
- The New Skill in AI is Not Prompting, It’s Context EngineeringPhilipp Schmid, June 2025.
- Marketing Tech Monitor 2026, Management SummaryMarketingTechLab, May 2026. 414 marketing decision-makers in Germany, Austria and Switzerland, January to February 2026.
- How can CEOs turn AI productivity gains into lasting value? (CEO Outlook, October 2026)EY-Parthenon, 1 October 2026. Anonymous online survey of 1,200 CEOs of large companies in 21 countries, August to September 2026.
Quotes are verbatim. Page numbers refer to the printed page of the PDF. Statements about the platform describe what it does today.
See it on your own brand.
14-day free trial. No credit card required.