Build your brand strategy with AI
Build a Brand Strategy With AI, but Keep the Choices Yours
AI tools sort a market's published claims quickly and can test a plan against them, and they write much the same plan for anyone who asks.
By Ines Calloway · November 17, 2026 · 8 min read

Working out how to build a brand strategy with AI starts with splitting the job. A model collects and sorts what your market already says, and it can test a draft plan against that material. Its own draft averages what's been published about your market, and a rival asking the same tool gets much the same draft.
Can AI create a brand strategy? It can produce a document with every heading a strategy usually has. What it can't do is make the three calls that make the strategy yours: which buyer you're for, which promise you can prove that a rival can't copy, and what you'll stop saying.
Sounding like everyone else usually comes out of careful research. Each team studies the same handful of competitors and answers the same template questions, and the answers converge. A model runs that benchmarking faster than any team, which gets a company to the spot its rivals already hold, sooner.
A model is fastest at sorting what your market already says
Before you ask a model for a single idea, give it what your market already says, starting with competitors' headline promises and sales lines, and your own customer notes. A model can collect and sort that material at a scale no person will sit through.
Melissa Kwan, co-founder of eWebinar, spent five months repositioning her company with Claude, and her account in Kyle Poyar's Growth Unhinged newsletter lists what she loaded into a Claude project first. The account is her own and covers one company, so read it as one founder's setup.
Her company sells on-demand video software with chat for demos, onboarding and training, to around 700 customers. For six years it called itself automated webinar software. Back on sales calls, she found that "Most people didn't fully understand what we did or what problem we solved."
Her project held call transcripts, more than 50 customer stories and 650 customer quotes gathered over six years, and a list of sales objections. She added the old website copy, kept as a record of what to move away from.
Her project also held a cheat sheet on each alternative buyers weigh eWebinar against. Some of those alternatives are habits, such as running the session live on Zoom. "Our competitors aren't other software products. They're substitute behaviors," she writes.
Last came the positioning she wanted to reach, and notes on the frameworks she trusts.
Whatever you sell, Kwan's setup translates into a short list. Gather each competitor's headline promises and sales lines, your own materials, what customers have said on calls or in reviews, and the objections your sales team hears.
Add a note on every alternative a buyer weighs you against, including the ways buyers handle the problem without buying anything.
Then have the model sort the pile. Ask which promises turn up across several competitors, and which words your customers use that no competitor's site does. The voice of customer research post covers that second list in depth.
Ask for several routes, because a model's ideas cluster together
With the material loaded, ask for several distinct strategic routes, and say outright that each one has to differ from the others: a different buyer, or a different promise at the front.

Left to its defaults, a model's ideas bunch up.
In a Wharton study on prompting for diverse ideas, Lennart Meincke, Ethan Mollick and Christian Terwiesch asked GPT-4 for consumer product ideas aimed at college students. They found "pools of ideas generated by GPT-4 with various plausible prompts are less diverse … than ideas generated by groups of human subjects."
They scored variety by how similar the ideas' wording was. Groups of people came in around 0.243, and GPT-4 ranged from 0.255 to 0.432, where lower means more varied. Even the model's most varied prompt produced ideas a little more alike than the people's.
One prompting method nearly matched the people. Chain-of-thought prompting, where the model is told to work through the task step by step, "was able to come close to what is achieved by groups of human subjects."
Consultants who used GPT-4 bunched up too. In a field experiment with 758 BCG consultants, those using GPT-4 to dream up a footwear product for niche markets showed "a marked reduction in the variability of these ideas compared to those not using AI." The authors suggest the tool "might lead to more homogenized outputs."
So ask for that step-by-step working explicitly, and point it at your market: have the model list what competitors already promise, then build each route away from that list, one at a time. Why that bunching matters more once polished copy costs nothing is covered in our post on brand in the age of AI.
The choices that make a strategy yours can't be handed to a model
Once the routes are on the table, the three calls from the top of this post are yours to make. Take the last one seriously: it covers lines you like that competitors also use.
A model works from consensus, and each of those decisions means stepping away from it.
Kwan drew the same line one level up. She picked her positioning method herself and used the model only to apply it.
"I asked AI to help me apply a method I had already chosen. It did not choose the method. Don't outsource your own thinking. The direction has to be yours," she writes. Elsewhere she puts it more bluntly: "Messaging fit is the founder's job."
For the buyer, start from the customer notes you loaded and ask whose problem you solve most clearly. For the promise, the test is proof a rival couldn't produce for themselves, and the guide to choosing a value proposition walks through that test.
What you'll stop saying comes straight from the sorting you asked for in the first step. Every promise that turned up on several competitors' sites is a candidate to cut. Your view of how your market works feeds all three decisions, and building a brand point of view has its own guide.
A route survives only if competitors don't already make its promise
Now run every route back through the competitive material you loaded. Ask the model, route by route, which of its promises competitors already make, and where. Then believe the answer, even about the line you liked best.

It's an awkward question to put to your favorite route. A model suits the job, though, because sorting what competitors say is the work it does fastest.
Say a made-up firm installs commercial kitchens, and one route leads on fast installation. If the model finds "fast installation" on most of the rival sites you loaded, that route goes, however good it sounded in the draft.
Keep pressing on the route that survives: ask what proof would back its promise, and whether any competitor could produce that proof too. The interchangeability test runs a similar check on copy you've already published.
Every conclusion the model hands you needs checking against the source
The consultant study carries a warning too. On tasks inside GPT-4's abilities, the tool helped. On a task that sat outside them, "consultants using AI were 19 percentage points less likely to produce correct solutions."
For brand work, the check is plain and tedious. When the model says competitors lead on price, open their pages and look. A customer quote it pulls should be findable, word for word, in a transcript you loaded.
Do the checking before a conclusion goes into the strategy, while dropping it is still cheap.
AI's ideas can outscore people's and still sound like a rival's
The strongest case for letting the model write the strategy also comes from Wharton. In a study of language models for idea generation, Karan Girotra, Lennart Meincke, Christian Terwiesch and Karl Ulrich had an online panel rate 200 student ideas and 200 GPT-4 ideas for consumer products.
"The average purchase probability of a human-generated idea is 40.4%, that of vanilla GPT-4 is 46.8%," they write. Of the top 40 ideas, 35 came from the model. If what you need is a long list of options buyers like, that's a real result, and the model wins it.
The same paper found the student ideas more novel on average. Its authors also say their analysis "does not speak to the similarity or variability in the content of ideas." The variety study, by two of the same authors, fills that gap: GPT-4's pools of ideas were less varied than people's.
An idea that buyers like but that sits close to what every other GPT-4 user gets still leaves you where your market already is. Both studies also tested consumer product ideas for students, some distance from choosing a position for a business.
Which result matters depends on your goals. For a quick pile of likeable ideas, take the model's. For a position a rival can't take, the model's first draft marks the spot to move away from.
If checking your routes against competitors leaves you with none standing, or two you can't choose between, we're happy to talk through what that means for your brand.
A model will read everything your market has published and hand the sorted pile to anyone who asks. Your rivals get the same pile, so the strategy only starts to differ at the point where you stop asking the model.
Frequently asked questions
- Can AI create a brand strategy?
- A model can produce a document with every heading a strategy usually has, and it is quick at sorting what a market already says. Three calls still have to come from you: which buyer you're for, which promise you can prove that a rival can't copy, and what you'll stop saying.
- How do you use AI for brand strategy?
- Load the model with competitors' headline promises and sales lines, your own materials, customer notes and sales objections before asking for ideas. Then ask for several distinct routes, test each one against what competitors already promise, and check every conclusion against the raw material.
- Why does AI make brands sound alike?
- In a Wharton study, pools of GPT-4 product ideas were less varied than ideas from groups of people, and in a field experiment with 758 BCG consultants, those using GPT-4 produced less varied ideas. A rival asking the same tool about the same market gets much the same draft.
- How do you get more varied ideas from an AI model?
- Ask for several distinct routes and say outright that each must differ, for example with a different buyer or a different promise at the front. In the Wharton study, chain-of-thought prompting, where the model works through the task step by step, came close to the variety of groups of people.
- Are AI-generated ideas better than people's?
- In a Wharton study of consumer product ideas, GPT-4's ideas had a higher average purchase probability than students' ideas, 46.8% against 40.4%, and 35 of the top 40 came from the model. The same paper found the student ideas more novel on average.
Sources
- How to build an AI brain for your positioning — Growth Unhinged, Oct 21, 2026
- Prompting Diverse Ideas: Increasing AI Idea Variance — Wharton Mack Institute, January 2024
- Navigating the Jagged Technological Frontier — MIT Sloan, Sept 2023
- Ideas Are Dimes a Dozen: Large Language Models for Idea Generation in Innovation — Wharton Mack Institute, Aug 2023