A Positioning Guide for AI Products

1. Competitive alternatives

Positioning starts with what your buyer does today, before they’ve heard of you. For most products, that alternative is software or people. With AI there’s a third bucket: the buyer tries the job in a general model, and you’re compared to what they’d get on their own.

Software that already exists

If the buyer already pays for software, you’re asking them to switch. Gamma competes with PowerPoint, Google Slides and Canva, so it has to show what it does that those tools can’t.

People or manual workflow

Some jobs are still done by people. A paralegal reviews the contract, and a support rep answers the ticket. If that’s the alternative, the buyer’s question is whether the work gets done as well. Investor Sarah Tavel has a simple test: if companies already pay outside firms to do the task, they’re ready to pay for the result.

DIY with a general model

The buyer can simply try the job in ChatGPT, Claude or Gemini. Harvey, which sells AI to law firms and corporate legal teams, faces this every day. Any lawyer can ask ChatGPT a legal question, but the answer can be confidently wrong, so Harvey sells the legal know-how that makes its output safe to put in front of a client.

2. Differentiated capabilities

Next is what you have that the alternative doesn’t. With AI, product performance is a weak differentiator. Your competitors build on the same models you do, so a new feature rarely stays yours for long.

For an AI company, the difference usually sits in one of the other nine, often in knowledge no model was trained on. Two contractors can buy the same tools at the same hardware store. You hire the one who knows your kind of house. As investor David Peterson puts it, “You haven’t eliminated specialization. You’ve moved it one layer up.”

Clay is a good example. Part of its edge is network, with more than 150 data providers and over a hundred agencies building on it. Its bigger move was on brand. Clay says it coined the term “GTM engineer” in 2023, which lets it define the job, and it teaches that job through its own courses, certification program and community. A competitor can copy its features, but it can’t easily take over the definition of a job Clay created.

3. Value

Capabilities are what you have, and value is what they change for the customer. Take each capability and ask “so what?” five times, the way you would in a 5 Whys exercise. For a tool that transcribes sales calls, it might look like this:

• It transcribes sales calls.
• So reps stop taking notes.
• So they give the buyer their full attention.
• So deals move faster.
• So each rep brings in more revenue.

With AI, the top of that chain has a shelf life. Stay too close to the feature and the next model copies it. Go too far and every sales tool can say the same thing. Stop at the first answer the budget owner would care about that still sounds like you. Here, that’s “deals move faster.”

4. Best-fit customers

Next is who cares most about that value. This is the choice HubSpot was making when it picked Mary. With AI, the tempting answer is everyone, because a general model can touch almost any workflow. But it’s rare to find the right market on the first try, and chasing every opening spreads a small team thin.

That’s why you should start with a beachhead segment: one small, well-defined group you commit to and win first, like taking one beach before moving inland. Sales, product and messaging all point at it. A good one can buy within a few months, pays enough to sustain the business, feels the problem strongly, and is within your reach. Teams that already buy AI tools often qualify, since they have the budget, have approved AI vendors before and know the output can be wrong.

Inside that beachhead, you go after what Maja Voje calls your early customer profile, or ECP. These are the customers who feel the problem badly enough to act now. They’ll try a product before it’s finished, put up with rough edges and tell their peers when it works. They’re rarely your biggest market, but they’re the fastest way to learn what the product is really worth.

Your ideal customer profile, or ICP, is who you build the business around. They get the most value over time, stay for years, spend more as the product matures, and make up a market large enough to carry the company. They also want proof before they buy, so you reach them by piling up pilots, case studies and references from your early customers.

With AI, expect the first pick to be wrong. Run small tests across a few promising segments and watch for early retention, expansion and willingness to pay before you commit. Most teams change their ICP at least once along the way.

Granola shows the sequence. Its beachhead was venture investors in Silicon Valley and London. They sit through a lot of meetings, like trying new tools, and talk to founders and other investors all day, so word spread fast. The product fit how they worked: no bot joining the call, just a notepad on their Mac that filled in the notes.

Then came the deliberate switch. At launch in May 2024, CEO Chris Pedregal said they were “done with VCs” and would build for founders instead, on the bet that a great product for founders would be a decent one for everyone else. Investors were the beachhead and the first early customers. Founders were the ICP. By October 2024, more than half of Granola’s users held leadership roles, though it had never set out to target executives. It was valued at $250 million in May 2025 and $1.5 billion in March 2026.

The hard part is turning away prospects who look like a fit but would need months of change before they see any value. You can come back to them once the product has caught up.

5. Market category

I think this is where being an AI company changes the most. The category is the frame buyers use to judge you, so pick the one that makes your value obvious to the customers who care most. The same laptop sold as a gaming machine gets judged on graphics. Sold to students, it gets judged on price and weight. Dunford describes three ways to play it.

Head-to-head

If a category is still up for grabs, you can fight for it directly. That’s rare in mature software and more common in AI, where many categories are only a few years old. AI coding is the clearest case. GitHub Copilot, Cursor and Claude Code are chasing the same developer budget, and each leans on a different strength. Copilot has GitHub’s reach, Cursor built its own editor, and Claude Code comes from the company that makes the model. It’s also the most expensive game to play.

Big fish, small pond

If a leader already owns the category, pick a segment it serves badly and win that. For most AI startups, I think this is the most realistic path, and it’s where your early customers come in. Attio did this in CRM. With Salesforce leading the category, Attio went after what it calls “GTM builders,” the technical founders and RevOps people who want to shape their own CRM. Many of its customers are AI startups, including Lovable and Granola.

New game

Creating a category is the hardest of the three, and it only works if the market needs the new one. With AI, it’s tempting to put “AI-native” in front of an old category name and call it new, but buyers still file you under the old category, so you’re compared to the same competitors for the same budget. Gong is a well-known example of a company that built a real category, with what it calls “revenue intelligence” in 2019. It now calls itself a “Revenue AI OS,” so it updated the label for AI and kept the category it built.

Relevant trends

The last piece is the trend itself, and here that’s AI. A CEO at an AI founder roundtable put the worry to Dunford: “AI capabilities are changing so quickly that any positioning we develop will be out of date in two weeks.” The fix is to re-run the steps whenever the models change. Most of the time, your value holds.

I’d add one caveat. Riding a trend in your copy can capture demand that’s there right now, as long as the position underneath stays tied to your value.

How much AI you put forward depends on your brand. If buyers in your category already know your name, AI can stay in the background. If they don’t, the category name and the AI label have to tell buyers what you are, and I’d guess that’s true for most companies reading this. Everyone in Montreal knows what Schwartz’s sells. A new deli on the Main has to put “smoked meat” on the sign. The copy on top will change more often than it used to, while the position underneath shouldn’t.

So, where should you start?

With your customers. Every step in this guide depends on what only they can tell you: what they’d do if you didn’t exist, and the words they use for what changed after they started.

That’s also how you make the bet more deliberate. The next model release will make some of your features common. The value your customers describe usually holds, so tie the position to it and re-run the steps when the models change.

I’ve worked through this in more than 40 positioning engagements. Getting to the value is the hard part, because customers rarely describe it the way you’d write it. Then comes the part most teams underestimate: getting sales, marketing and product to use that language in every conversation.

None of this is a formula, and two people can read the same market differently. The steps are quick to run. Getting each answer right takes judgment, mostly about what to leave out.