Brand strategy for 2027: what 2026 is teaching us

Imagine asking an AI assistant to choose a brand for you. Not to admire its campaign or appreciate its carefully renovated typeface, but to find something that does the job. The assistant returns three names and a paragraph. Somewhere inside that paragraph, years of marketing expenditure have been reduced to a few claims.

As we look towards 2027, this is the uncomfortable question for brands: what remains when someone else does the summarising? If the answer is “quality, innovation and a commitment to customers”, the problem is not the algorithm. The problem is that you have described almost everybody.

The wrong response is to produce more material for the machine. The better response is to become clearer about what deserves to survive the compression: a useful difference, credible proof and a promise the organisation can actually keep.

My thought as we look towards the year ahead is this: let AI lighten the work without losing the judgment that gives it meaning. The research emerging in 2026 offers a useful starting point for thinking about that balance. Here, I explore what we can learn from it and what it might mean for brands preparing for 2027.

The new shop window is someone else’s answer

In research published in October, McKinsey reported that 53% of surveyed consumers, on average, had used AI for a shopping task in the previous three months; among AI shoppers, 81% agreed that it helped narrow and filter choices (McKinsey).¹ These are survey responses, not independently observed purchases, but they point to a consequential change: a brand can be assessed before its own website enters the conversation.

That shifts the practical question from “How do we get people to our content?” to “What account of us will people encounter elsewhere?” The two belong together. Neither makes the other obsolete.

For 2027, I expect more brand work to involve checking those accounts: whether an assistant understands the product, confuses it with a competitor, repeats an outdated claim or recommends it for something it cannot do. This is less glamorous than launching another campaign. It is also closer to the customer’s decision.

Begin with the questions a buyer actually asks. Which option suits a small team? What works without specialist support? What will this cost after the introductory offer? If your material answers none of these clearly, the missing ingredient is not another acronym for search optimisation.

Make the product understandable. Keep prices, limitations, evidence and service details current. A machine-readable brand should first be a readable one.

Do not budget for an automatic challenger advantage

There is an appealing story in which AI strips away advertising muscle and recommends whichever product best fits a need. It is particularly appealing if you are not the category leader. But a new interface does not necessarily mean a level playing field.

A November preprint audited 19,200 answers across three engines and four sectors using Italian queries, finding a positive relationship between brand popularity and mentions in 23 of 24 tested configurations (The Popular Get More Popular).² It is an early, context-specific study, not a universal law, but it challenges the comforting idea that a conversational interface starts the market again from zero.

My prediction is that established recognition will remain an advantage in 2027. A smaller brand’s response should be specificity, not an expensive attempt to sound as large and vague as everybody else.

Consider an illustrative software business. “Helping organisations transform” gives a buyer almost nothing to work with. “Helping regional distributors reconcile orders without replacing their existing systems” identifies a customer, a problem and a constraint. It also gives the business a proposition it can support or disprove.

Earn recognition around a concrete use. Publish the evidence behind it. Give customers language they can repeat without borrowing your pitch deck. This is slower than manufacturing a library of generic articles, but it creates something those articles cannot substitute for: a reason to remember you.

Efficiency is not the same as distinction

Imagine removing the logos from six AI-assisted campaigns. Could anyone reliably tell which belonged to which brand? If the answer is no, reducing the cost of production has not solved the most important creative problem.

An older experiment remains useful here. In 2024, researchers Anil Doshi and Oliver Hauser found that AI-generated ideas improved evaluations of individual short stories while making AI-assisted stories more similar to one another (Science Advances).³ This was a writing experiment, not a commercial branding study; the distinction matters, but so does the possibility it exposes.

Something can improve on its own and become less distinctive in company. Better sentences do not necessarily produce a more recognisable voice.

For 2027, I expect the practical advantage to move towards better inputs and more demanding editorial judgment. Your useful inputs are the things a generic prompt cannot supply: an awkward customer question, a local detail, a product trade-off, a founder’s actual conviction or a specific observation from frontline staff.

Give the machine something worth working with. Then ask more of the result than grammatical correctness.

Would a competitor comfortably publish it? Has it explained anything the reader did not already know? Does it contain a claim somebody could challenge? These are inexpensive questions. They are also rather less flattering than a dashboard announcing that the team has increased output by 300%.

Customers want help, not compulsory delegation

The temptation in a technology forecast is to turn every emerging behaviour into an inevitable destination. People use AI to compare products; therefore, people will soon want AI to manage the entire relationship. That second proposition requires its own evidence.

Kantar’s October research found that 32% of consumers researched brands or products through AI assistants, but only 23% trusted their brand recommendations (Media Reactions 2026).⁴ Its November US shopping article reported that only 28% were comfortable with AI making a purchase on their behalf, while nearly half were very comfortable with it automatically applying deals (Kantar).⁵

The studies do not share an identical scope, and the November article omits detailed sample information. Still, the strategic distinction is useful: permission to assist is not permission to act.

In 2027, I expect brands to discover that the most welcome automation is not always the most ambitious. Find the discount. Explain the difference. Remember the preference. Ask before spending the money.

Build around graduated permission rather than maximum autonomy. Let customers inspect a recommendation, correct an assumption and reach somebody when the situation stops fitting the script.

The competitive question is not “How much of the customer journey can we automate?” It is “Where does automation make this easier without taking away control?” One starts with the organisation’s appetite. The other starts with the person paying.

Authenticity is a question of responsibility

An automated stock update and an automated apology should not be treated as the same communication problem. One conveys information. The other asks a person to believe that an organisation understands what happened and means what it says.

In seven preregistered experiments, Colleen Kirk and Julian Givi reported reduced positive word of mouth and loyalty when emotional marketing communications were believed to be AI-written; the effect weakened in contexts including factual messages and AI editing rather than authorship (Journal of Business Research abstract).⁶ This does not establish that every AI-assisted message damages trust. It does suggest that what AI replaces matters.

A September 2026 conceptual paper makes a related argument: AI disclosure can change how people interpret the way an advertisement was produced, and what that implies about the advertiser (Journal of Advertising).⁷ It is a framework, not another experiment.

My expectation for 2027 is that broad assurances about “human involvement” will become less useful than specific explanations. What did the system do? What did a person check? Who can correct the result?

Do not create a performance of human care around a process nobody owns. Use AI for drafting where it helps, but preserve meaningful judgment where the message claims attention, understanding or commitment. An apology does not become sincere because somebody approved its punctuation.

The brand guidelines need to leave the folder

A brand book can describe a voice without explaining how to make a difficult decision. That is a weakness worth fixing before giving an automated system more room to operate. “Warm, confident and human” is not an answer to whether it may promise a refund.

The World Federation of Advertisers’ September study found that 96% of the major brand owners surveyed used generative or agentic AI, while only 8% had a clear AI-discoverability strategy (WFA).⁸ Its sample was small and concentrated among large advertisers, so it should not stand in for every business. It nevertheless illustrates how adoption can outrun clarity.

In 2027, I expect useful brand guidelines to become more operational. They should say which claims require evidence, what information must not enter a tool, which decisions need approval and what happens when an answer is wrong.

This is where brand strategy meets internal communication. Employees need examples, practice and permission to question an output. A policy that cannot be understood during a busy working day is not much of a safeguard.

Start with ten real situations rather than 100 abstract principles. Show an acceptable answer, a misleading one and the reason for the distinction. Name the person responsible for resolving uncertainty. The aim is not to turn every employee into a compliance officer; it is to stop making confidence the only qualification for pressing publish.

Do not let imaginary customers approve the strategy

An audience that is always available, inexpensive to consult and willing to answer every question has obvious attractions. Before replacing customers with synthetic versions, however, ask whether convenience is being mistaken for insight.

A 2026 study compared GPT-4o-generated demographic “twins” with 461 US soft-drink consumers and found that synthetic responses overstated positive brand attitudes and showed less variation than human responses (NIM Marketing Intelligence Review).⁹ The average attitude difference was 1.2 points on a seven-point scale, a result limited to that category, sample and model rather than a verdict on every synthetic research method (NIM study).

For 2027, my recommendation is to use synthetic audiences to improve the questions, not certify the answers. Generate possible objections. Rehearse an interview. Look for assumptions that need testing. Then speak to people before committing to consequential positioning decisions.

Keep synthetic material clearly labelled and separate from actual testimony. An invented quotation does not become customer evidence because it sounds plausible in a presentation.

Most importantly, retain the inconvenient responses. The person who misunderstands your offer, rejects the premise or notices something your team has stopped seeing may be the most useful participant in the room. A brand should not automate away its opportunity to be surprised.

India needs a more specific conversation

“Indian consumers are ready for AI” is the kind of sentence that makes a presentation easier and a strategy less precise. Ready for which task, in which language, with what risk and what route back to a person? Those are the questions worth funding.

Ipsos’s 2026 India report found that more than 90% of respondents wanted organisations to be open about when and how AI was used, but its sample comprised 1,000 affluent online respondents in NCCS A, a socioeconomic classification (Ipsos India report).¹⁰ It is evidence about that audience, not permission to flatten a country into one adoption statistic.

My prediction is that useful local understanding will matter more than generic declarations of AI leadership in 2027. The test is whether the experience works for the intended customer, not whether the English-language demonstration looked impressive.

For a brand serving several language groups, investigate comprehension rather than treating translation as completion. Can customers understand the offer, the disclosure, the limitation and the way to challenge a mistake? These are research questions, not outcomes established by the Ipsos report.

Make your claim at the level you can support. If an AI feature helps a particular group complete a particular task, explain that benefit. There is no need to recruit an entire nation into your product announcement.

Being mentioned is not the same as being chosen

A brand can appear in an AI answer and still lose the sale for perfectly sensible reasons. The price is wrong, the service is unavailable, the recommendation misreads the need. Visibility deserves attention, but treating it as the final outcome turns measurement into another form of self-congratulation.

As you plan for 2027, build the scorecard around three separate questions. Did the brand appear? Was it described accurately? Did the encounter help someone make a decision? Keep the answers separate, because a rise in the first can conceal a failure in the other two.

Use a small, documented question set rather than screenshots of especially pleasing answers. Include comparisons, constraints and reasons not to buy. Repeat the exercise over time and record the tool, date and wording, so that a changed answer does not immediately become evidence of a changed market.

Then connect the exercise to actual customer research. Ask buyers what they consulted, what confused them and which claims they checked. Treat those recollections as clues rather than perfect records of influence. A person’s route to a purchase is not always something they can reconstruct neatly for your attribution report.

The same discipline applies to content. Count production savings, certainly, but also check whether the material improves understanding or reduces avoidable questions. An article that attracts attention by exaggerating a capability may improve one chart while creating work for sales and support.

My forecast is that better teams will become less impressed by isolated AI metrics and more demanding about their meaning. The useful distinction is between evidence that the system can find you and evidence that a customer has a good reason to choose you. Do not buy a dashboard that encourages you to forget the difference.

What to do now, and what to plan for early 2027

My strongest prediction for 2027 is not technological. It is that the gap between fluent promises and lived experience will remain a better place to look for brand problems than the number of tools a team has bought.

Preparing your brand strategy for 2027 should therefore begin now, with a small set of uncomfortable checks. Use the remaining weeks of 2026 to define the promise, identify the evidence behind it and decide who owns it. You do not need to wait for January to remove a claim nobody can substantiate.

For January 2027, plan a closer examination of your five most important promises. Ask real customers what they understand each one to mean, and compare those interpretations with what the organisation can deliver. Use the discrepancies to choose what to fix before deciding what to produce next.

For February, schedule a review of representation and experience together. Repeat a documented set of buyer questions across relevant AI tools, then compare the answers with product information and actual service processes. A correct description of a bad experience is not a marketing success.

For March, set aside time to test a few changes with real people. Choose one clearer product explanation, one more distinctive creative approach and one service interaction with explicit customer control. Measure understanding and task completion alongside speed and cost.

Resist turning this into a single “AI readiness” score. The point is to discover what needs attention, not produce a flattering average.

The work is making the promise usable

The boundary between brand and organisation is less useful than it looks on a department chart. A promise needs people who understand it, processes that support it and communication that makes sense when something goes wrong. Otherwise, a more efficient content system merely distributes the contradiction faster.

This is the space my work connects: AI enablement, communication and brand identity. I help teams turn complex ideas into clear messages and practical ways of working, rather than treating the rollout as the finish line.

If your team is preparing for 2027, a useful starting brief is concrete: one confused message, one uncertain workflow or one customer promise that has become difficult to keep. From there, we can work on the communication, guidance and practical support needed to make it usable.

The question is not whether your brand can generate more. It is whether someone can still tell what you stand for, why it matters and who will take responsibility. Start with that, and the technology has a job worth doing.

Research notes

Written on 12 November 2026, this article examines 2026 developments to date, not a completed year. All references to 2027 are forecasts, recommendations or proposed plans, not observations of events that have happened. Two earlier papers are retained to explain mechanisms relevant to the current debate.

1. McKinsey, “Who wins when AI does the shopping? New evidence on agentic commerce”, 28 October 2026. Approximately 21,000 consumers across 11 markets; fieldwork in June 2026. Self-reported behaviour and attitudes, with bases varying by question.
2. “The Popular Get More Popular: Auditing Brand Bias in Generative Search”, Giorgio Monaco, Jacopo Lenti and Francesco Pierri, arXiv preprint, 5 November 2026. Italian-query audit across three engines and four sectors, plus controlled ranking tests. Findings are context-specific and should not be treated as peer-reviewed consensus.
3. Anil R Doshi and Oliver P Hauser, “Generative AI enhances individual creativity but reduces the collective diversity of novel content”, Science Advances, 12 July 2024. Randomised study involving 293 UK writers and 600 evaluators. Short-fiction results are not direct evidence of commercial brand effects.
4. Kantar, “Net 75% of marketers plan to increase investment in AI visibility, including ads” (Media Reactions 2026 press release), 2 October 2026. Over 23,000 consumers and 800 senior marketers. The release omits detailed fieldwork and sampling information.
5. Kantar, “Holiday 2026: Shoppers trust AI to find deals, but not to make purchases”, 9 November 2026. US shopping survey synthesis. Detailed sample sizes and methods are not provided in the article.
6. Colleen P Kirk and Julian Givi, “The AI-authorship effect”, Journal of Business Research, 2025. Seven preregistered experiments. Findings and moderators were checked against the accessible abstract; full experimental tables were not reviewed.
7. Denitsa Dineva and Oguz A Acar, “Generative AI Disclosure in Advertising: From Intent- to Production-Based Disclosure”, Journal of Advertising, published online 24 September 2026. Conceptual framework rather than original experimental evidence.
8. World Federation of Advertisers, “96% of major brands now use AI in marketing”, 30 September 2026. Based on 54 senior respondents from 46 major brand owners, not a representative sample of all businesses.
9. Kaiser and colleagues, “Leaving Insight to Digital Twins? Promise, Progress and Limits of Synthetic Respondents”, NIM Marketing Intelligence Review, 2026. Comparison of 461 US soft-drink consumers with GPT-4o demographic twins. Single-category and single-model scope limits generalisation.
10. Ipsos, “India’s AI Moment: Confidence, Optimism, and Pragmatism”, AI Monitor 2026 India Report. Online sample of 1,000 NCCS A respondents. Results should not be generalised to India’s entire population.