A founder can run 30 days of marketing with AI by choosing one buyer problem and one primary channel, then repeating a weekly cycle: speak to prospects, publish one useful piece, send a permission-based email, make a few relevant outreach attempts, and review results. Use ChatGPT or Claude to draft and organise; keep positioning, claims, relationships and spending decisions human.
The plan works best as a test of a specific offer, not as a substitute for knowing customers. If no one understands the product after a clear demonstration, more AI posts will multiply confusion. Reserve time each week to hear objections, update the offer and stop messages that misdescribe what the startup can actually deliver.
Day 1: choose the buyer and the proof you can show
Write a one-page offer brief. Include who has the problem, what event makes the problem urgent, how they handle it today, what the startup does, what it does not do, the price or enquiry route if known, and the evidence available now. Distinguish a tested result from a founder belief. If the product is not live, say so. The first marketing goal might be five useful discovery conversations rather than immediate sales. A plan that asks for purchases before the offer can be explained will hide the real gap.
For an illustrative specialist food importer startup, the first offer could be a small catalogue of hard-to-source pantry ingredients for independent kitchens. The buyer is a purchasing manager who needs reliable availability and clear pack details. The current proof is a verified supplier list, sample pack specifications and a delivery process, not a claim that the startup already saves every customer money. The founder chooses the buyer's question “Can I get the same ingredient consistently with usable specifications?” as the month's content centre.
Use a simple evidence ledger. Each proposed claim has a source and an approval status: “pack size 2 kg — supplier specification, verified”; “dispatch in 24 hours — not yet proven”; “saves 20% — unsupported”. The ledger keeps the AI assistant from turning aspirations into facts. The marketing claim evidence tutorial gives a deeper check; for this month, the founder needs only enough structure to stop an untrue message before it goes live.
Choose one primary place where the buyer already seeks information: a relevant professional community, a search-oriented resource page, a trade newsletter or direct conversations. Keep one owned follow-up route such as an opted-in email list. Do not call all channels “equal experiments”; each needs time to learn. A kitchen fitter selling to homeowners would choose a different primary channel from an ingredient importer selling to purchasing teams. The principle is to start where the buyer's decision begins.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowDays 2–4: set up one source of truth for drafts
Create a small project in ChatGPT or Claude with the offer brief, approved facts, audience notes, voice guidance and a weekly task list. Current vendor help confirms both products have projects for keeping related instructions and source files together, subject to account or workspace settings. Do not upload confidential supplier terms or customer correspondence unless the chosen account and permissions permit it. A project is a drafting workspace, not an automatic CRM or publishing system. If a co-founder or part-time helper needs the same workspace, check sharing before you choose: OpenAI's help lists project sharing on ChatGPT plans from Free upwards, with chat or edit access per person, while Anthropic's help says Claude projects can be shared with colleagues on Team or Enterprise plans. On an individual Claude plan, each person keeps their own copy of the brief, which is exactly how two versions of the evidence ledger start to drift.
Give it instructions that fit a founder's month: “Use only approved claims; mark unknowns; propose one useful action per message; ask before asserting a delivery promise; write in plain language; produce a short review checklist with each draft.” If the project cannot reliably maintain a fact across chats, paste the current brief into the prompt again. Do not rely on model memory for a price, product availability or date that can change during the month.
Act as a drafting assistant for a founder-run 30-day test.
Use only the offer brief and evidence ledger below.
For each draft, show the buyer problem, proof used, call to action,
and any claim that needs verification. Do not invent customer
results, names, quotes or stock. Keep outreach under 120 words.
Ask one clarifying question when a fact needed for publication
is missing.
An illustrative output might say “trusted by hundreds of kitchens” after reading that the founder previously worked in food supply. Those are different facts. Delete the customer-count claim. A better line is “We are testing a catalogue with independent kitchens and can send current specifications for the ingredients you use.” The founder should check whether even that testing statement is true. AI is useful for options; the founder is the source of accountability.
Set up a basic tracker with date, channel, asset, intended buyer, claim check, owner approval, link, reach, meaningful replies, enquiries, sales or next step, and learning. A spreadsheet is enough. Do not buy an analytics suite before you can record what was sent and who responded. The preparation tutorial helps organise inputs if your documents are scattered.
Days 5–7: interview before committing to a content calendar
Invite five plausible buyers to short conversations. State that you want to understand their process, not pitch them into a script. Ask what they currently buy, how they check specifications, what goes wrong, who approves a new supplier and what evidence would make a trial worthwhile. Take notes with consent and remove personal details before sharing them with an AI tool. A founder can ask AI to cluster themes after the calls, but should reread the raw notes to catch disagreement.
An illustrative first interview says, “I care less about ingredient novelty than whether a replacement will have the same allergens.” Another says, “I can test a sample, but I need a predictable pack size.” The tempting AI summary might be “buyers want unique products”. That misses the operational problem. The founder corrects the theme to “specification consistency and predictable ordering” and changes the first content piece accordingly. A single striking quote is not market proof, but it can expose a weak assumption.
If five conversations are impossible in the first week, use the best available real evidence: replies to relevant enquiries, support questions, publicly asked questions and observations from the founder's own sales calls. Label which evidence is direct and which is inferred. Do not manufacture a customer persona with demographic details that no one has supplied. The month can still proceed, but the day-30 decision should acknowledge the thin evidence.
Week 2: publish one anchor piece and three useful extracts
Write one substantial piece that answers a buyer question with a practical method. For the illustrative importer, “How to compare two ingredient specifications before replacing a supplier” could include a short checklist: ingredient composition, allergens, pack size, storage, lead time and change notice. It must rely on documents the founder can verify. AI may produce an outline, turn notes into a first draft and suggest clearer headings. The founder checks each technical statement and removes anything that implies regulatory certification not established.
Turn the anchor into three distinct extracts rather than repeating the same introduction. One could be a short checklist for a professional post. Another could show an anonymised comparison table with illustrative data. A third could answer an objection heard in an interview: “Why a cheaper unit price can still create more purchasing work.” Link back to the full piece where that is useful. If the founder uses a social channel, respond to genuine questions rather than only broadcasting scheduled content.
A sample prompt is: “From the approved supplier-comparison piece, draft three posts: one checklist, one before-and-after process, and one answer to the pack-size objection. Each must teach a distinct point and avoid claims about our delivery speed. List which source lines each post uses.” An illustrative output may invent a “48-hour guarantee” because it sounds persuasive. Remove it and replace it with “ask for the current lead time before ordering”. The source ledger should make that correction straightforward.
The founder can spend an illustrative three hours on the anchor: one hour drafting with AI, one hour checking sources and rewriting, and one hour formatting and publishing. The three extracts may take another hour to adapt and review. Those timings are planning assumptions. If checking the claims takes four hours, record that; a technical content bottleneck may be important evidence about where the business needs better source documentation.
Do not measure success only by impressions. Record whether the right buyers asked a more specific question, downloaded a specification, requested a sample or booked a conversation. A post with 2,000 views from irrelevant readers may be less useful than one reply from a purchasing manager. The marketing measurement tutorial explains how to connect activity with meaningful outcomes.
Week 3: send a permission-based email and careful outreach
Write one email to people who actually asked to hear from the startup. It should share the anchor piece or a helpful excerpt, explain why it matters and invite a clear next action. It does not need a countdown or a discount. Send a test copy, open every link and check it on a phone. If the list is tiny, the founder can still learn from replies. A small list is not permission to add every contact from an old address book.
An illustrative email says: “Hello [first name], several purchasing teams told us supplier changes are hard to evaluate when the pack and allergen details arrive in different formats. We made a one-page comparison checklist. You can use it with any supplier. If you would like current specifications for the ingredients we carry, reply with the product you need and I will send the right sheet. [your name].” It offers value before asking for a conversation and avoids a fabricated testimonial.
For direct outreach, select ten organisations with a clear reason to contact them. Record the public evidence for fit, the appropriate contact route and what you can offer now. AI can draft a first version and check length, but the founder should verify the recipient, the specific observation and the tone. Do not infer private buying problems from a person's profile. Follow the channel's rules and stop when someone declines. A good message leaves the recipient a simple way to say no.
A before-and-after example shows the difference. Weak: “I noticed your company needs better ingredients; our revolutionary AI-powered sourcing changes everything.” The founder has no basis for the need or the claim. Better: “I saw that your menu includes several oat-based items. We are testing a small supplier catalogue and can send current oat specifications and pack options if that would help your next comparison.” The latter still needs a real observation and a truthful offer. The AI has cleaned the writing only after the founder supplied evidence.
Book a human follow-up for interested replies. Do not let an automated message promise stock or price without checking it. If a prospect asks for a sample, confirm availability, address handling and terms through the normal business process. The lead-generation agent tutorial can help later when the founder knows what a qualified lead is; month one is for learning that definition firsthand.
Week 4: repeat the useful action, then assess the evidence
Use questions from week three to choose the second anchor piece. If buyers ask about pack sizes, publish a comparison that makes pack options clear. If they ask about allergens, produce a source-checked explanation of how the business supplies specifications, without offering regulatory advice beyond your competence. AI can cluster questions and propose an outline, but the founder chooses the question that affects a buying decision. Reusing the original theme without a new answer would be easy content and weak learning.
Send a second email only if it gives subscribers something different: a new checklist, a clarified product update or an answer to a frequent objection. Make a few additional outreach attempts informed by what you learned, not by simply increasing volume. Review any ad experiment against a capped budget and a working destination. A founder without a marketing hire should protect time: a content schedule that consumes every evening and produces no useful conversations is not sustainable.
An illustrative failure occurs when the founder queues 30 AI posts on day two. On day ten, an interview reveals that customers care about pack consistency rather than novelty, but the queue continues pushing “discover exciting ingredients”. Pause the queue, correct the message and record the wasted work. Scheduling is useful after a clear buyer problem has been validated. In a fast-learning startup, human feedback should be able to interrupt automation.
The 30-day numbers for one founder
For the illustrative food importer, assume the founder can devote six hours a week, or 24 hours across the month. Week one uses six hours for the offer brief, project, tracker and five conversations. Week two uses six hours for one source-checked anchor piece, three extracts and responses. Week three uses six hours for one email, ten researched outreach messages and follow-ups. Week four uses six hours for a second anchor, a second email and the review. These are budgeted hours, not a claim that every founder works at that pace.
Suppose the founder sends ten personal outreach messages in week three and eight in week four. Five recipients reply, three ask for specifications, two request samples and one starts a paid trial order. The first email goes to 70 opted-in readers and produces nine clicks and two useful replies; the second goes to 74 and produces six clicks and one reply. The two anchor pieces together bring 160 measured visits. Do not add replies, sample requests and orders as if they were independent customers; the same person may appear in several stages.
Assume the founder spends $0 on paid ads, uses an existing AI account and pays an illustrative $20 for a separate email or design expense during the month. The cash expense is $20 plus any existing subscriptions and product samples; the meaningful cost is also 24 founder hours. If the single paid trial order contributes an illustrative $45 after product and fulfilment costs, that does not prove a scalable marketing return. It is one signal. The founder should ask whether the trial buyer matches the target, whether fulfilment worked and what would make a repeat order likely.
Compare those results with a prewritten day-30 threshold. For example, continue the same buyer and channel if at least three real conversations reveal a recurring problem, at least two qualified prospects request specifics, and the founder can fulfil the promise. If replies show a different problem, revise the offer and repeat a narrower test. If outreach produces no relevant replies despite checked fit and clear wording, inspect the audience or channel before producing more posts. These thresholds are illustrative management rules, not industry benchmarks.
Keep four decisions with the founder
The founder owns positioning: which problem is worth solving and what the product can honestly do. They own claims: every number, customer quote, delivery promise and comparison needs evidence. They own relationships: an outreach note should reflect a real reason for contact, and a reply should answer the person's actual question. They own spend: an AI suggestion to increase ad budget is not approval. Delegating draft production can create time for these decisions; delegating the decisions themselves can make a young business sound certain before it has learned.
At day 30, keep the assets that survived contact with buyers: the offer brief, evidence ledger, two useful pieces, approved email templates, response notes and a simple results table. Delete or archive drafts that made unsupported claims. Update the project sources so the next month's AI drafts reflect what you learned. If marketing is only one of the jobs you want AI to carry, the 90-day startup AI playbook places this month inside a wider plan for choosing which business jobs AI takes on first, and which stay manual. The plan is working when the founder can say, with evidence, who responded, what they needed, which message helped and what the next small test should be.
Founder-run marketing plan questions
How many channels should a founder use in the first month?
Start with one primary channel where likely buyers already spend attention, plus one way to collect permission for follow-up. A second channel is reasonable only if the founder can maintain it and measure a distinct result. Reusing a good idea in another format is fine; opening five accounts and posting thin variations is not a useful test.
Can AI send outreach messages automatically?
Use AI to organise research and draft messages, then have the founder check the recipient, factual premise and offer before sending. Respect the channel's rules and the recipient's preferences. For a first month, a small number of relevant human-approved messages gives better learning than an unreviewed automated sequence.
What if the startup has no customer stories yet?
Describe the problem, product demonstration, founder's reasoning and questions from real discovery conversations without inventing a customer result. You may label a hypothetical example as illustrative. Ask early users for permission before quoting their words or showing their work. AI can organise evidence, but it cannot create genuine proof of demand.
Further reads
- Can AI Write a Marketing Plan for a Small Business? — Compare a longer planning process.
- Can a One-Person Business Automate Its Marketing With AI? — Choose tasks a solo operator can sustain.
- How to Test a New Service Idea With AI Before You Launch It — Test a new offer before expanding promotion.
- AI Marketing Tools vs Hiring a Marketer: A Cost Comparison — Compare ongoing founder effort with paid help.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI help, Projects in ChatGPT; Anthropic help, How can I create and manage projects? Both read November 2026. The plan timings and example numbers are illustrative.