How to Build a Go-to-Market Strategy with AI: From Market Research to a 90-Day GTM Plan
This is the second part. In the first — State of Go-to-Market 2026: what works, what died, and how AI changed everything — I covered what changed in the market. Here I cover how to work under those conditions: the methodology itself and where AI sits inside it.
Executive Summary
In May I ran a six-week master group on building a market entry strategy. Six two-hour sessions, a lot of homework, and detailed reviews of each participant's work. It was the first cohort where my GTM methodology and AI worked together.
The methodology rests on fifteen years of practice: close to 200 projects across nine countries, first at Soloviov Marketing Agency, which later became Soloviov Consulting. It came together as a system over the last three years, when I started teaching it and had to show people how to do the same thing on their own. While you work alone, a lot of it rests on intuition; the moment you teach someone else, you are forced to formulate the rules.
I ran the GTM Master Group in 2023 and 2024 as well, with 150+ participants in total. In 2026 I rebuilt the frame so the model sits inside the methodology rather than next to it.
Can Claude or ChatGPT build a market entry strategy for you? Short answer: no. And that is exactly the point. But it can help a great deal, and that is what this article is about.
A Go-to-Market strategy, as I understand it, is a system for entering a market: who we sell to, why they would buy from us, through which channels, on what sales logic, and against which metrics. More importantly, it is not just a set of actions but the order in which they are carried out — and I will come back to that, because in practice the sequence matters to models too. Material has to be given to them gradually, in that same order.
I use several models in my work: Perplexity and Gemini for search, ChatGPT for processing large data sets, Claude for complex intellectual work. Most of my time is spent in Claude.
AI will process thousands of sources, generate hypotheses and structure research in minutes. But a GTM strategy is a sequence of decisions: which market we enter, who we sell to, why they would buy from us, which channel we put money into, what we test first, and what data would make us change our minds.
Those decisions are made by a person. The model helps prepare them faster.
Important: I don't use Claude to generate GTM strategies. I use it purely to speed up parts of a methodology I developed myself and have already validated across nine markets.
In the previous report I showed the figure: 91% of B2B companies adopted AI, and most see no result. There I covered why, and what is broken. Here I cover what context to give the model so it processes what you actually need. I will show how I do it.
- how I assemble the context layer for working with Claude,
- which nine blocks make up my methodology for building a market entry strategy,
- how market research and customer profiling are done with AI,
- which decisions I never delegate to the model,
- the signs by which I recognise AI slop and
- how many hours cleaning it up actually costs, and
- how all of it comes together into a 90-day GTM plan.
01. Why prompts don't produce a strategy
A document that looked finished
Back to the master group that ran in May. The homework after the first session was simple: fill in the GTM frame for your own business.
At the second session one participant, the founder of a design agency, brought his. He had done it properly: for months he had been building his own knowledge base about the agency inside Claude, uploaded industry research into it, and asked the model to pull everything into a structure.
The document came out senior. Positioning, competitors, market, annual goals, sales channels. Not superficial — with numbers, with references to research. It looked like the work of someone who understands their business well.
One block in it was empty in substance while being full in volume. The block about customers.
He knew everything about business planning in general and almost nothing about his buyers specifically. That data simply wasn't in the knowledge base. Claude didn't say "I'm missing information." It filled the block in as confidently as everything else.
AI doesn't know what it doesn't know. It fills gaps in context with plausible text — and by eye you can't tell it apart from the rest of the document.
Two weeks later the same participant rewrote that block from scratch. Not with the model — from his own experience and from conversations. He ended up with four distinct customer profiles instead of one blurred description. And everything downstream changed: communication, product line, call scripts. He stopped offering early-stage startups a large site with branding and started selling them what they actually need at that stage. The first sale into the new segment happened while the programme was still running.
When unlimited access doesn't help
The second case is in the same category, but more expensive.
From March to June I was preparing a large strategy session for a client together with partners: 18 senior executives in the room, each with their own agenda. We ran several stages of research, and customer development interviews were one of them. For the research part I brought in a contractor — a strong specialist with unlimited access to a top model.
Every time, he produced a volume that was impossible to work with. Each week a new mass of material, formally correct and impossible to make a decision from. We went round in circles for almost three months. It was only resolved when I asked him to put the model aside and reassemble the picture by hand, from what he had actually heard in the interviews.
Unlimited access to a powerful model doesn't give you a better result. It gives you more material. And more material is not the same thing as more clarity.
The formula everything starts from
Both cases are described by one equation:
AI result = model + context + task.
Most people work only with the third term. They look for the perfect prompt, adjust the wording, rewrite the task. And every time they start from a blank sheet, because the model knows nothing about their business.
GTM doesn't work that way. A market entry strategy requires persistent context: the business, the product, the market, the ICP, the personas, the competitors, positioning, pricing, channels, and the history of what has already been tried and what came of it. Without that layer, any prompt produces a generic answer, however well it is written.
02. The context layer: how I train Claude for a project
Here is how it works for me. This is my working standard, not a universal recommendation — I tried different ways of organising it and settled on this one because it holds up across projects.
A trained funnel
For each project I create a separate conversation window in Claude and teach it step by step. Everything goes in: industry research, my own materials and frameworks, competitor data, interview transcripts, previous campaigns and their results.
After a few weeks of that work it is no longer a chat. It is an environment that knows the project context and the way I think about it. When I come in with a task like "assemble the messaging for this segment", I don't have to explain who the client is, what the product does, or what we already tried in the spring.
That was the goal of my programme: for every participant to leave not only with a finished strategy but with a trained funnel that holds the whole methodology along with the context of their business.
That, honestly, surprised me more than anything else. A methodology I had spent years on turned out to be transferable in a way that works without me.
Artifacts: five documents that hold the system together
The second element of the context layer I brought in not from AI practice but from donor-funded projects — the NGO world taught me to build these artifacts long before the models arrived.
An artifact is a document built to parameters I define, not the model. I have several dozen of them, but five hold the system together.
- 01
Customer profile
The central document in the whole system. Mine is already built and worked through in detail: not a sketch like "business owner, 35–45", but specific triggers, objections, decision criteria, and the language people actually use.
- 02
Research
A set of prompts and checklists I load into Claude so it runs the research and pulls out exactly what is needed rather than everything at once.
- 03
Materials, articles and lectures
My own texts, teaching materials and frameworks, uploaded into the model. That way it gets access to what I know, not only to what is on the open web.
- 04
Value proposition framework
Built on top of the customer profile. The formula is fixed: what we offer, to whom, which problem we close, by what means, what measurable result the client gets, and how we differ.
- 05
Channel marketing plan
A scenario document adapted to the specific project, its capabilities and its budget. It breaks down as needed: PR, performance, outreach, partnerships, events.
Artifacts are convenient precisely because they are easy to load back into the funnel. You get a closed loop: context produces documents, documents deepen context.
03. What a market entry strategy is made of
Where this methodology came from
The methodology for building a market entry strategy did not appear alongside AI, and it was not written as a methodology. It formed over years of practice — agency projects, product launches, work with donors and with startups.
It started turning into a system when I had to hand it to someone. In 2023 and 2024 I ran cohorts of the master group, and teaching forces you to make explicit what you had been doing on instinct.
That matters practically. When a methodology existed before the models, you can see clearly what the models speed up and what they don't do at all.
The nine GTM blocks
My market entry strategy consists of nine blocks. It is a working document, filled in over several passes: first the hypotheses, then the evidence, then the corrections.
- 01
Target audience
Who the customer is, what hurts, who makes the decision, who blocks it, and which trigger sets the purchase in motion. The most important block, and the one most often filled in with guesswork.
- 02
Value proposition
What we offer, to whom exactly, which problem we close, by what means, what measurable result the client gets, and how we differ from the alternatives.
- 03
Competitors
The list, what each does well, and your position relative to them. It pays to go beyond desk research and collect real signals from the market.
- 04
Market analysis and windows of opportunity
Size, trends, speed of entry — and, separately, the cultural context of the buyer. More on that below, because this is where AI misleads most often.
- 05
Marketing goals
In numbers and in defined terms. Not "increase awareness" but specific revenue, a number of deals, or qualified conversations by a given date.
- 06
Sales strategy
How the sale actually happens: direct, through webinars and workshops, through partners, a referral programme, a free product, or a combination.
- 07
Channels
Where this audience already exists today. Not where you would like to be, but where it currently reads, asks questions and makes decisions.
- 08
Budget
Broken down by channel, with a separate portion for experiments that will almost certainly fail. That portion has to be allocated deliberately.
- 09
Metrics and timelines
Plan versus actual for every channel. Without it you will not learn what works, and you will scale blind.
Sequence matters more than the blocks
The nine points above are easy to find in any textbook. What works is the order in which they are filled in.
Market research and trends → target audience → value proposition and messaging → channels → marketing plan → execution and measurement.
Each block feeds on the one before it. You cannot pull one out of context.
The most typical violation I saw in the programme. One participant, the owner of a SaaS service, set out to write the value proposition without having worked through the customer profile. Everything he produced came down to "cheaper than competitors."
And worse than that — "cheaper" repels a B2B buyer, because it reads as "lower quality." He couldn't formulate value not because he wrote badly, but because he had not closed the block above.
A weak value proposition almost always means an unclosed block further upstream.
From my own practice, and from the audits I have run for clients, a properly worked value proposition reduces cost per lead and shortens the sales cycle more reliably than any change of creative.
04. Market research and ICP with AI
How not to do it, and how to do it
A typical research request looks like this: "Claude, research the Spanish market for me." What comes back is a tidy overview in which everything is true and nothing is usable. It is an encyclopedia entry, not evidence you can act on.
Guided research looks different. It is a chain in which every link is set by you:
question → sources we trust → evidence → synthesis → hypotheses → gaps in the data → next research questions.
And a second rule I apply to any research output: separate fact, model guess, hypothesis and decision.
How to define an ICP: from market to purchase criteria
An ICP (Ideal Customer Profile) is a description of the company and the person who need your product most and are able to pay for it.
The path from a broad market to a working profile runs through seven steps, and AI is useful at almost every one:
Broad market → segments → ICP hypotheses → who uses it and who pays → purchase trigger → pain and objections → selection criteria.
Claude structures segments well, generates hypotheses, and challenges your assumptions if you ask it to directly. It does one thing badly: it does not know which segment is worth your money.
So the final ICP is not generated. It is chosen. It is a strategic decision about which part of the market you are deliberately giving up.
Processing interviews and large data sets
A separate note on where the model gives the biggest speed gain. I regularly process product interviews through Claude to extract the exact language people use to describe their problem — that language then goes straight into messaging.
The same applies to large-scale research. When you have several hundred interviews or a big sociological data set, processing it by hand within a reasonable timeframe is impossible. Here the model genuinely replaces weeks of work.
05. Positioning and messaging: where the model turns generic
At the third session we worked on value propositions. Two of the three participants brought versions generated by the model. I rejected both with one word: generic.
The writing was smooth. The problem was that those formulations could have been placed under any company in the same niche and nobody would have noticed the swap. That does not work.
The third participant got it right on the first attempt. The difference was not in the prompt and not in the model. He was the only one who had built the customer profile before writing the text.
Generic output almost always means generic input. The model has nothing to be specific with if you didn't give it specifics.
The chain where meaning gets lost
Messaging is a chain. The audience, its pains, the frames those pains produce, and then the website, the ad accounts, the call scripts, the proposal.
Something is lost at every handover. Most often it is lost at the first one: the person doesn't know their audience's pains precisely enough and compensates with general words.
Different people hear different things
The second thing people constantly skip: the message has to be separate for each segment and for each role in the decision.
The CTO asks about data security, where it is stored, and how fast the integration is. The commercial director asks about price and payback period. The CEO asks about business risk.
When the message is the same for everyone, it lands with no one. That is where the rising cost per lead comes from.
What the model genuinely does well here
Positioning is chosen, not generated. It is a decision about what you want to occupy in the customer's mind and what you give up in order to do it. But within that decision the model is useful:
- Alternatives
- Five positioning options for the same product, so it becomes clear your first version was not the only one.
- Uniqueness check
- The question "could this be said by any of our competitors?" — asked of every line.
- Weak points
- Where the claim is unsupported and where it will be challenged.
- Objections
- Simulating the reaction of each decision-making role.
- Reconciliation with the profile
- Whether the message actually matches the pains recorded in the ICP.
06. Can AI build a GTM strategy
No — and the reason is not that the models aren't smart enough. In each of the nine blocks there is a part the model does and a part the person does.
- Market
- The strategist chooses where we go. Claude gathers evidence and synthesises sources.
- ICP
- The strategist chooses the priority segment. Claude structures hypotheses and challenges assumptions.
- Positioning
- The strategist decides what we want to occupy in the customer's mind. Claude generates alternatives and simulates objections.
- Messaging
- The strategist approves the key message. Claude tests variants across segments and channels.
- Channels
- The strategist sets priorities and allocates budget. Claude finds where the audience is already present.
- Sales
- The strategist interprets real conversations. Claude finds the repeating patterns in them.
- Plan
- The strategist makes the decisions and the deliberate trade-offs. Claude structures and operationalises.
Six decisions I never delegate
- Choice of market — where the company's limited resources go.
- ICP prioritisation — who we genuinely focus on, and who we walk away from.
- Positioning — what exactly we want to occupy.
- Resource allocation — where we put time and money.
- Strategic trade-offs — what we deliberately decide not to do.
- The final content of the 90-day go-to-market plan.
The reason is simple: for each of those decisions I am accountable with money. The model is accountable for nothing.
AI accelerates hypothesis testing. It does not test the market for you.
The limit you see in customer interviews
Another participant ran around ten in-depth interviews with his audience. The respondents were delighted. One even took a screenshot of the product to show a colleague. Then came the question of paying, and it turned out that none of them were ready to buy at the price the business model required.
No model would have shown that. Willingness to pay is not derived from market analysis, from a description of competitors, or from segment size. It is established in conversation.
The second mistake in the same case turned out to be more interesting. Countries for the interviews were chosen at random — whoever replied got interviewed. As a result most respondents came from markets that were never the commercial target.
07. The check: AI slop and what it really costs
There is a stage in the methodology that appears in no framework and takes up more of my time than any other. Checking for AI slop.
I check almost everything. Research I do myself. Material I receive from contractors. Text that goes to a client. Clients spot AI slop instantly, and it costs trust rather than time.
Part of it I catch simply through exposure: you see that something is off and go and verify it. But exposure isn't transferable, and you have to work with contractors somehow.
And separately, the part that never shows up in a budget. What takes most of my time is not producing material but correcting other people's. When a contractor sends over generated text, the hours go into the checking, not the writing.
The term "AI slop" describes the situation where somebody generated something somewhere, and then a whole department spends weeks verifying it and bringing it into usable shape.
Four signs I recognise it by
- 01
Repeating words and constructions
The same turns of phrase migrate from paragraph to paragraph. The eye catches it before the brain finishes reading the content.
- 02
Built through negation
"This isn't just a tool, it's a solution." "Not because of X, but because of Y." One such phrase is fine. When there are five on a page, a machine wrote the text.
- 03
An excess of em dashes
The most visible technical sign. Live people don't punctuate that way.
- 04
Text you take nothing away from
The main sign. You finish the paragraph and know nothing more than before. There is no sense of a person having been involved: no observation of their own, no number, no position you could argue with.
Part of this can be removed at the input stage
I keep a separate instruction document for Claude that removes most of these signs at the generation stage. It sets out how not to write like a machine: which constructions to avoid, which rhythm to hold, where a specific is required instead of a generalisation.
The document is not my invention. Instructions of this kind are based on the list of signs of machine writing compiled by Wikipedia editors in the article Signs of AI writing.
Two details make it genuinely work, and both are easy to miss.
08. Entering a new market: market fit and channels
One strategy for all markets doesn't work
I was building a Nordic market entry strategy for a client. The market was researched, the segments defined, the messages ready. Then I ended up at an industry event and got a picture that didn't match the desk research at all.
Half of what we had built had to be thrown out. In the Nordics the B2B deal cycle runs up to twelve months, and trust is built through personal connections rather than through a strong offer.
Same product, same ICP, same messages — and a completely different sales motion. Claude won't tell you that, because it isn't in its context until you put it there.
Market fit: the product doesn't land in every market
- Nordics
- The person who sells is called Johansen and grills at another Johansen's summer house. Trust is built over years through personal ties, and the deal cycle is long.
- US
- Speed and a clear number. The buyer wants to understand the value in the first minutes of the conversation.
- EU
- Process, compliance, and several decision-makers with different criteria.
That is market fit in the practical sense: not "is the product needed there" but "can we sell there the way people there are used to buying." A question about behaviour, not about demand.
AI SDR: six months, zero deals
A foreign marketing agency working with clients in Southern Europe and Israel launched an AI SDR and tested it for six months. Zero closed deals.
Over the past year the same story repeated on several B2B SaaS projects entering Poland, Canada, and a group of five markets across Europe and the US. The repetition matters more than any single case.
The failure mechanism is identical. The tool parses websites and profiles, generates "personalised" emails, and sends them out. The result is generic at industrial scale.
Where your audience already is
The question the channel block starts from sounds like this: where does this audience already exist today. Not where you want to be, but where it currently reads, asks and decides. Communities, partners, events, outreach, content, the founder's network, referrals, and AI search.
I run lead generation as one system rather than a set of separate channels: outreach, Meta performance, Google Ads, and visibility in AI search. Channels have to reinforce each other. Outreach prepares the ground for a conference, a webinar feeds the funnel, a partner closes the deal because they already have the customer's trust.
AI search as a stage of the sale, not just a channel
The best example here is my own, so I will use the numbers from Soloviov Consulting.
In the previous report I wrote about the first inbound leads from AI search. There are now seven. These are people who came neither from advertising nor from cold email: they asked a model who understands market entry, and the model named me, specifically on a query about building market entry strategies.
One of those calls stayed with me. A director from a state university phoned, asked very specific questions, and had clearly prepared. She had found me through a model and arrived already knowing what she wanted.
More important than the leads themselves is a second observation: practically all my clients check me through ChatGPT before first contact. Not after the meeting — before it.
That changes the status of AI visibility. It has become a stage of the sale. The question is no longer whether ChatGPT will bring you a client, but what it says about you when a client asks.
I tested this on myself at one point and got an unpleasant result. The model described work I had done three to five years earlier and positioning I had already moved away from.
The model describes you through your old content. If you changed your positioning and the content stayed as it was, in the AI's answer you are still the person you used to be.
09. The 90-day GTM plan
All nine blocks converge into a single document, and it is not a strategy presentation. It is a 90-day plan in which you can see where the leads will come from, at what cost, and by when.
The most frequent mistake at this stage is substituting an operational plan for a marketing one. People have brought me excellent documents: preparation phases, responsibility matrices, ordered processes, nested stages. It looks solid. But a document like that answers the question "how is our operations set up", not the question "where will the leads come from".
If an action doesn't convert into an inbound lead, it doesn't belong in the marketing plan. That is a different document.
And the second rule I repeat on every project: a strategy differs from an illusion by having a plan for reaching the goals. No plan with numbers and dates — no strategy.
What a 90-day plan should contain
- A list of channels, each on its own line.
- Budget per channel, separately.
- Plan and actual for leads from each channel.
- An owner for each channel.
- Start and end dates for every action.
- A tracking tag, set before launch rather than after.
- 15–20% of budget for retargeting, warming up those who aren't ready yet.
- At least three channels that reinforce one another.
What it looks like in practice
Take a real case from the programme. A design agency, effectively no budget for paid traffic, audience of early-stage startups.
We open the customer profile and look at where that audience already is. Accelerators and entrepreneur support programmes. There are five or six of them in the market, all well known. From there the plan writes itself.
Five talks in five communities per quarter. Roughly a hundred listeners. After each talk, a free review as the entry point. A realistic conversion into conversations, and from those into deals.
Nothing technological about it. But it is a plan in which every action leads to a meeting, and in which there is a number you can check against reality a quarter later. Claude helped assemble it faster; the decision about where to go was made by a person.
Sales conversations as data for strategy
The last element, the one that closes the loop. Twenty sales conversations are a data set most companies never use at all.
Transcripts of those conversations go into the context layer, and from there Claude does a good job of extracting recurring objections, frequent questions, selection criteria, and the exact words people use to describe their problem.
Conversations with customers become training data for your GTM system. Not for the model — for the strategy.
What comes out of it
We worked through the methodology described above over six weeks on real businesses: a B2B service, a SaaS product, and an education platform.
Results appeared before the programme finished. One participant entered a new segment and closed the first sale there. Another signed a contract they had been circling for months.
The methodology works, but it is strict about sequence. Skip a block, or take it out of order, and the result falls apart.
Context → Research → Hypothesis → Test → Conclusion → Adaptation → Scale.
And an honest list of what AI does not do, however much context you load into it. It does not guarantee demand for your product. It does not replace conversations with customers. It does not take responsibility for the choice of market.
What interests me most is the point where strategy meets first revenue. A market entry plan is worth very little if nobody owns the messaging, the pipeline and the follow-through.
I am currently packaging this methodology into a GTM playbook: nine blocks, artifact templates, and a prompt library for each block. I will write about the launch in my Telegram channel.
Frequently asked questions
Can Claude build a GTM strategy?
No. A GTM strategy is a choice of market, priority segment and positioning — and that belongs to a person, because a person carries responsibility for the result and for the money. Claude accelerates research, structures hypotheses and tests wording, but it does not make the choice.
How can AI be used to develop a market entry strategy?
As a multi-layered field and a working environment for research, analysis and execution inside a structured methodology — not as a generator of finished strategies. The model works well when it has persistent project context: the customer profile, research, competitors, previous campaigns.
What is AI-powered GTM?
It is a market entry strategy in which human decisions are combined with AI support at the stages of research, hypothesis generation, analysis of feedback and preparation of materials. The decisions stay with the person; the speed comes from the model.
Can AI replace a GTM strategist?
Definitely not. AI accelerates part of the GTM work, but it does not replace understanding of the customer, prioritisation, or accountability for the result. And the main thing: willingness to pay is established in conversation, not derived from analysis.
What should a 90-day GTM plan include?
ICP, positioning, messaging for each segment, priority channels, experiments, budget by channel, metrics with plan and actual, owners, and dates. Every action in it has to convert into an inbound lead.
How long does it take to develop a market entry strategy?
Developing the strategy and handing it to the team takes 2–4 weeks. If it needs to be driven through to first sales, that is a fractional GTM lead engagement running several months.