Can AI do your startup’s marketing? A founder’s weekly workflow with a time budget

What does AI do well in marketing today?

Anything where a good first draft saves you from a blank page, and a person still checks the result.

The honest split is between producing and deciding. Current assistants are strong producers. Given clear context, they write a usable first draft of an email, a landing page section or a post in seconds. They summarise a call transcript into the three points that matter. They turn one long piece into several short ones. They research a list of companies, pull out the public facts and put them in a table. They read a spreadsheet of last week’s numbers and describe what changed.

The tasks worth handing over first, because they are frequent, bounded and easy to check.

• Summarising sales and customer calls into buyer language, objections and questions.
• First drafts of posts, emails and page sections from a brief you wrote.
• Repurposing: a talk into posts, a post into an email, a long page into a short one.
• Account research: what a target company does, recent news, likely priorities.
• Weekly reporting: turning exported numbers into a short summary with the counts attached.
• Variations: five subject lines or three openings to choose from, not to send all of.

Notice the verb in each line. Summarise, draft, repurpose, research, turn into. Each ends with something you read before it goes anywhere.

What still needs the founder?

The decisions, the relationships and the responsibility for what is claimed.

AI does not know your buyer unless you tell it, and it cannot find out by talking to one. It will write confidently about a customer it has never met. It cannot pick your channel, because it does not know where your buyers actually spend their time; it knows where buyers in general are said to. It does not carry your reputation. If a draft overstates what your product does or invents a statistic, the damage lands on your name, not the tool’s.

Keep these with you, whatever the tools can do.

• Positioning: who you are for, what you replace, and why now.
• Buyer conversations, and the judgement about what they mean.
• Choosing the one channel you will work this quarter.
• Every factual claim, number, customer name and comparison.
• Relationships: replies, introductions, partners, the people who refer you.
• The final read of anything published or sent.

Volume is the specific trap. It is now easy to publish a great deal, and more output feels like more marketing. It is not the same as finding customers. Search engines have said as much: Google Search Central’s guidance on AI-generated content (updated December 2025) allows AI-assisted content but warns that using such tools to generate many pages without adding value for users may breach its spam policy on scaled content abuse. The weekly workflow below is deliberately small for that reason.

What does the weekly workflow look like?

Seven slots, under five hours, the same every week.

The time boxes below are a budget, not a measurement. They are what the work should take once the context file exists and the routine is familiar. The first two weeks will run longer. Keep a note of actual time against each box, because the gap between the two is the most useful number this workflow produces.

That adds up to four hours and fifty minutes. The Wednesday slot is the one founders drop first, and it is the one that makes the others work. The call gives Monday its angles, Tuesday its language and Thursday its proof. Without it, the tools are drafting from their own general idea of your buyer, which is exactly the content the scaled-content warning describes.

If you have read our note on doing marketing with no team, this is the same routine with the tools slotted in. One number, one thing shipped, one conversation and one thing stopped. AI shortens the drafting and the reading. It does not add slots.

What should you give the AI so it stops guessing?

One document, written once, updated every Friday. It is the difference between your voice and everyone’s.

Most weak AI marketing is weak because of what went in, not because of the model. A tool asked to “write a LinkedIn post about our product” fills every gap with the average of everything it has read. A tool given your buyer, their words and your proof has far less to invent.

Keep a single context file with six headings.

• The buyer: one paragraph on who they are and the moment they start looking.
• Their words: phrases lifted from calls, updated after every conversation.
• Claims you can prove: each with its source, such as a customer quote you have permission to use or a documented product capability.
• Claims you never make: promised results, invented numbers, comparisons you cannot support.
• Voice: three short examples of writing you are proud of, and five words you never use.
• This quarter: the one channel, the one offer, the one number you read.

Then ask for work in the same shape every time.

• Context: paste or attach the context file.
• Task: what to produce, for which channel, at what length.
• Point: the single thing the reader should take away.
• Constraints: use only claims from the provable list; do not invent figures, names or quotes.
• Flags: list every factual claim in the draft and where it came from, and mark anything you were unsure of.

The last line is the useful one. Asking the tool to list its own claims turns the approval step from a close reading into a checklist.

What is the approval rule, and why does it matter?

Nothing leaves under your name until you have read it, checked it and would say it out loud to a customer.

Four checks, in order, before anything is published or sent.

• Read all of it, not the first paragraph. Tools can drift in the middle.
• Trace every claim, number and name to the provable list or a source you can open. Remove what you cannot trace.
• Say it aloud. If you would not say it to a customer on a call, rewrite it.
• Check the ask. One clear next step, and one that you can actually handle if people take it.

Early on, do not connect any tool that can publish or send without you pressing the button. Scheduling a post you already approved is fine. Letting a tool decide what goes out, or reply on your behalf, removes the one step that protects your reputation, and the time it saves is small next to the cost of one wrong claim to the wrong buyer. The machine drafts, a person decides.

The machine drafts, a person decides.

When should you stop doing this yourself?

When the time box keeps breaking for good reasons, not bad ones.

Founder plus AI is a stage, not a destination. The log you keep every Friday tells you when it is ending.

Signs it is time to bring in help.

• A channel is producing conversations and you cannot keep up with the follow-up.
• The workflow overruns every week because there is more worth doing, not because drafts need rescuing.
• You have started skipping the Wednesday conversation to make room for production.
• Sales or product work is visibly slipping to protect the marketing hours.

If it overruns for bad reasons, such as drafts that need rewriting from scratch, fix the context file before you hire. A person you bring in will need the same file, so the work is not wasted. When you are ready, our scorecard for the first marketing move after a seed round sets out how to choose between a head of marketing, a fractional CMO and an agency, and our note on marketing agencies versus AI marketing tools covers what each one actually sells.