How to Position an AI Product That Does Too Much

You write a headline about one thing your product does well. It reads clear. Almost suspiciously clear.

Then you remember the other things it does. The reporting took a month. The agency use case could be big. Someone at a demo mentioned recruiting, and the product could probably do that too.

You add a sentence. Broaden the audience. Put "platform" back in.

By the end the homepage describes an AI-powered workspace for modern teams. Everything you built is represented. The person reading it has to work out what they would use it for.

I did exactly this to GrowthMentor's homepage more than once. The product does a lot. Mentors, help requests, roadmaps, call recordings, a matching engine. Every version of the page that tried to say all of it said nothing. The version that works today opens with one comparison, GrowthMentor against the AI you already use, because that is the alternative every founder was weighing when they landed on it. Not MentorCruise. ChatGPT.

That is the whole method in one example. Pick the alternative the buyer is really comparing you with, and write for that.

Mine went the same way before it went back. Each pass added a noun and lost a reader.

The narrower sentence feels like a smaller company

There is a reason "an AI platform for marketing teams" is easier to approve than "get your next batch of product ads ready for review". The first has room for everything on the roadmap. The second sounds like a tool with a fairly ordinary job.

After all the work you have put in, ordinary can feel insulting.

You know about the model orchestration, the integrations, the parts you rebuilt until they worked. A buyer sees a way to get some ads reviewed. Same product, very different reasons for valuing it, and the buyer's version is the one that fits in a budget line.

I read the requests that come into GrowthMentor. In the four months to September, founders describing an AI product used the word "platform" 87 times across 590 distinct requests, "all-in-one" 20 times, and "AI-powered" 25 times. The word "wrapper" appears zero times. Nobody thinks their product is the generic one. The buyer, reading five of these homepages in a row, cannot tell them apart.

The awkward part is that you often can serve the second audience without much extra development. Why choose ecommerce marketers if agencies could use it? Why stop at ads if the same system writes email?

Because supporting another output and understanding another buyer's purchase are different amounts of work. An agency needs client approvals and separate brand accounts. A solo store owner wants someone to decide what to advertise. An email generator answers neither.

Keep the possibilities. Ship a starting offer a customer can understand without hearing the expansion plan. Leaving something off the homepage does not delete it from the product.

Pick the customer who cares about the difference

"Be more specific" sends people the wrong way. You add a company size, an industry and a job title until the customer description sounds precise, and you still do not know why that person would buy.

Specificity is only useful when it explains a purchase.

A marketer who occasionally needs a rough draft is fine with a general-purpose tool. A marketer pushing recurring batches through brand review cares about approved inputs, revision history and claim checking. The same generation feature serves both. The rest of your product is only worth paying for to the second one.

Start there. Which buyer gets a real advantage from the parts of your product that are different?

Look at the evidence you already have. Who brought their own material to a demo? Who keeps using the same capability? Who asked detailed questions about fitting it into their week? Those people are worth understanding. A compliment about the product's potential tells you almost nothing.

If you have no evidence yet, choose a buyer you can reach whose job the current product already handles, write down why you think they would switch, and treat that as a hypothesis in your ideal customer profile. You are allowed to be wrong about the first customer. Keeping every possible customer in the pitch is what stops you finding out where.

Give their current approach some credit

"Couldn't I do this in ChatGPT?" sounds dismissive when you have spent months building, and the instinct is to answer with everything the buyer missed in the demo.

First find out what they mean by "this".

If they mean generating a rough draft, their current tool is probably enough, and I have told founders on calls to walk away from a product that was competing with a model vendor's native features. If they mean getting a batch approved, exported and into a campaign, there is more work in it than they think. Ask them to show you the last attempt. Look at the preparation and the corrections, not only the output. The full version of that question is in the 9 AM test, and this article is about what you say once you have passed it.

April Dunford's approach to competitive alternatives starts with what the customer would do if your product did not exist. Manual work, an existing tool, or living with the problem. She also warns against treating every company you find online as an alternative your buyer considers.

You may be watching another startup's launch while your prospect decides whether to keep paying their freelancer.

That freelancer has an advantage worth taking seriously. The buyer sends a brief and someone takes responsibility for the result. Your software might generate faster and still leave the buyer with more work, because now they operate it and fix the output.

An incumbent tool has a different advantage. It is already part of the team's day. Replacing one step means moving files, approving another vendor, and teaching colleagues where the work lives now.

Those costs belong in your comparison. A product has to be enough better at something the buyer cares about to make the change worth it. And if the buyer is happy with their current process, accept that they may be a poor fit. A lot of copywriting effort goes into making people dissatisfied with a tool they like.

Make a promise that can be challenged

Take an illustrative AI creative product that generates images, writes copy and produces campaign variations. Its broad pitch is "create high-performing content at scale".

That sounds ambitious. It also lets almost any output count as progress.

Now choose ecommerce marketers who already have a brief and approved product assets, and struggle to get batches of ad variations through review. Suppose the product generates the variations, checks copy against supplied product information, and keeps feedback attached to each version.

A possible opening would be this.

Get your next batch of product ads ready for review. Turn an approved brief and product assets into ad variations, check the copy against your product information, and collect revisions before export. Built for ecommerce marketers running regular creative tests.

You can hear the objections already. What if they do not have approved assets? What if writing the brief is the hard part? What about a brand whose real problem is deciding which campaign to run?

Good. Now you can see who the offer does not help.

The broad version let all those buyers imagine something different. Five encouraging conversations, five incompatible product requests. The narrow version gives you a chance to find several buyers who want the same thing, and something specific to dispute. A marketer might tell you collecting feedback is easy and correcting the generated images takes forever. That is useful. You will not learn it by debating whether "intelligent" sounds better than "AI-powered".

Notice the limit of the promise. Ready for review says nothing about whether the ads sell. To claim performance you need campaign evidence. To claim less review work you need to compare the full process, corrections included, with what the buyer does now. Write the promise at the level you can demonstrate.

Here are three more rewrites of the kind that happen on calls here. Each is fictional. Each follows a pattern I have watched mentors run on real homepages, and the most common move in all of them is the same. The word AI leaves the headline and reappears further down, where it explains how the work gets done.

Bring the broad headline and the narrow one

A mentor who rewrites AI product positioning on calls every week will tell you in thirty minutes which buyer the narrow one is for, and what the broad one is hiding. Unlimited calls, one membership.

Talk to a mentor

Put AI where it answers a question

There is no prize for removing "AI" from every sentence. A buyer looking for an AI coding assistant needs to recognise the category. A technical buyer may need to understand the model and its constraints before anything else.

The question founders bring to calls is narrower than "should I say AI". It is whether a specific framing helps or hurts. Whether "AI data agent for ecommerce" converts better than "one place for your store's numbers". Whether "AI consulting" reads as a category or as noise to a business owner who has heard it forty times this quarter. In the four months to September, mentors moved AI out of the headline on roughly twenty calls, and in almost every case put it back lower on the page.

For the creative product, the buyer needs to know what is generated, what the system checks, and what a person still reviews. The label explains how the work gets done. It does not explain why this product fits their work better than the alternatives.

Try reading your headline with the word AI removed. If what remains is "better content, faster", you still owe the buyer an explanation. Name the work they can hand over and the work they keep.

That last part matters if your team is still behind the scenes repairing output. A service that delivers a usable batch is a perfectly sensible offer. Describe what the customer receives and who does the work. Calling it self-serve software creates an expectation the experience cannot meet.

Let the demo expose the limits

Once you have chosen a promise it gets harder to hide behind an impressive demo, because the buyer has something to check.

This is less comfortable than running your favourite example. It is closer to the job you are asking someone to pay for.

Google's People + AI Guidebook recommends helping users understand capabilities and limits so they know when to apply their own judgement, and explaining relevant data sources and how to recover from errors. Apply that to the claim you are making. "Checks product claims" needs an explanation of what it can check, what it misses, and who approves the result. A badge will not answer those questions.

If you have no customer results yet, work through representative material with a prospective buyer. Agree on what a useful result looks like before running it. A failure here is uncomfortable and it tells you what to change. Narrow the promise, improve the product, or try a buyer whose requirements you can meet. Any of those beats rewriting the sentence again.

Test the position without rescuing every conversation

A broad pitch gives you room to improvise. The buyer looks bored, you show another feature. They mention a different use case, you explain how you could support it. With a flexible product you may even be right.

You finish the call having agreed on something entirely different from what you started selling, and you learn nothing about whether the original offer works.

For the next round of conversations keep the buyer and the promised job consistent. Same opening in outreach, homepage and demo. Keep a note of four things.

• What they expected. Ask what they thought the product would help them do. A mismatch is a communication problem you can fix.
• What they use now. The process and why they keep it. Convenience, trust, or the job is not painful enough.
• Where they stopped believing you. The claim, the missing capability, the unanswered question. Another adjective will not fix a failure in the demo.
• What happened next. Material for a test, the approver joined, a pilot, or a no. Keep the reason next to the outcome.

Let a poor-fit buyer stay a poor fit. If someone needs a service and you are testing software, write that down and investigate the service offer separately. Patterns across the round decide the next change. The buyer understands the offer and does not need it, or wants it and cannot trust the output. Those are different decisions. Broadening the headline treats them as the same one.

Decide what you will leave out

Before the next homepage revision, finish the sentence from the top of this page in plain words. It can be clumsy. You are making a decision, and elegant wording can wait.

The second list is harder to write than the sentence. It holds work you are proud of, markets you have not explored, and the reasons you thought the company could be big. None of that disappears. It has to stop competing with the customer's reason to buy today.

If you keep putting everything back, that is the moment to bring both lists to someone who has done this pick with money on the line. I still do it with mine. The people below read segment lists for a living, and the useful answer to "which omitted capability would change this buyer's decision" is usually none of them.

The outside read

Mentors who rewrite AI product positioning on calls every week, and will tell you which list is right:

Dani Whitestone

Growth strategist. Fixes the message before the funnel. 100+ sessions.

Book a call

Roisin Bennett

AI go-to-market strategist, positioning and product marketing, twenty years in. 90+ sessions.

Book a call

Daniel Johnson

GTM operator, ten years scaling AI and B2B SaaS startups. 400+ sessions.

Book a call

Then give the narrower version a fair test. A buyer should be able to turn you down for a reason you understand. That is progress over needing another demo to explain what you meant.