Content
AI Marketing for Small Business: What Actually Works
5 min read
AI made content production cheap. It did not make attention cheap.
That gap is where most small business marketing now fails. You can produce thirty posts a month for almost nothing, and thirty posts a month that nobody reads is worth exactly what you paid for them. The bottleneck moved — from writing to being found, and from being found to being trusted — and most advice hasn’t caught up.
Here’s what actually moves a number when you’re one person with a limited budget.
The three uses that pay back
1. Making one good thing, several times
The highest-return marketing activity for a small business is producing one genuinely good piece of work per week, then adapting it for every surface it belongs on. Not copy-pasting — adapting: a different opening for each platform, a different length, a different ending.
This used to take a day per piece. It now takes forty minutes. That’s the real gain, and it compounds: four weeks of consistent, adapted output beats four months of sporadic original output, every time.
The failure mode is producing thirty mediocre pieces instead of adapting one good one. If the underlying piece isn’t worth adapting, more distribution won’t save it.
2. Answering the questions people actually type
Every business has a set of questions customers ask before buying. Those questions are the highest-intent search traffic available, and almost nobody answers them in public.
AI is genuinely good at this job: take twenty real customer questions, group them by what the asker was trying to do, and produce a structure for answering each one properly. You supply the experience; it supplies the organisation.
What it can’t supply is the answer. If you let it write the substance, you get a page that says nothing, and a page that says nothing ranks for nothing — or ranks briefly and then doesn’t.
3. Testing what you’d never have tested
Small businesses don’t test because tests feel expensive. Ten headline variants, three email openings, five subject lines — these used to be a week of work. Now it’s twenty minutes.
The discipline that matters: pick one variable, run it long enough to mean something, and write down the result. Most people generate variants and then pick the one they liked best, which is not a test, it’s a preference.
The three uses that waste time
1. Volume without a distribution plan
If you can’t say where a piece of content will be seen and by whom, producing it is a hobby. This is the single most common failure, and AI makes it worse because production is now frictionless.
Before producing anything, answer: who will see this, how will they get there, and what will they do next? If any answer is vague, the content is premature.
2. Personalisation that isn’t personal
“Hi [First Name]” is not personalisation. Neither is an email that references the page you visited. Buyers recognise both as automation and discount them accordingly.
Real personalisation requires knowing something specific, which requires attention, which doesn’t scale. At small-business volume, one genuinely specific email to twenty people beats a personalised-looking email to two thousand.
3. Chasing the platform of the month
Every quarter there’s a new channel that’s “the” opportunity. For a one-person business, the cost of learning a new channel is two months of not doing the current one well.
The test: have you already exhausted the channel where your buyers already are? If not, the new channel is avoidance dressed as strategy.
What actually works, in order
For a small business under $50k a month, this sequence beats any tool stack:
- Make the offer legible. A stranger should understand what you sell, who it’s for and what it costs within five seconds. Most small business sites fail this, and no amount of traffic fixes it.
- Answer the buying questions in public. Ten pages that answer real pre-purchase questions, each pointing at the relevant product.
- Get in front of people who already have the problem. Communities, directories, newsletters, other people’s audiences. This is unglamorous and it works.
- Follow up with people who showed interest. One email, one question, no sequence. Most small businesses never do this and it’s the cheapest revenue available.
- Then consider paid acquisition, once the above converts organically.
AI helps at steps 1, 2 and 4. It doesn’t help at step 3, which is the one that requires you to talk to people.
A two-week plan that fits in the gaps
Not a strategy. A schedule for someone who has two hours a week for marketing.
Week one, hour one: write down the twenty questions customers ask before buying. Verbatim, in their words. This is the raw material for everything else and it takes forty minutes if you have a support inbox.
Week one, hour two: pick the three questions that come up most. Write one page answering each — properly, with the actual answer, not a summary that points elsewhere.
Week two, hour one: turn each page into three adapted pieces for wherever your buyers actually spend time. Different openings, different lengths. One product link per piece, at the end.
Week two, hour two: find three places where people with this problem already gather. Post one of the pieces in each, as a contribution rather than an advert. Answer every reply.
That’s six pieces of content and three distribution actions in two weeks, from two hours a week. It’s not impressive. It’s repeatable, which is the only property that matters.
What to measure when you’re small
Not impressions. Not follower growth. Not engagement rate.
Three numbers, checked monthly:
- How many people arrived from search? This is the only traffic that compounds.
- How many of them asked a question or bought? Traffic that doesn’t act is a vanity metric with better branding.
- Which page did it? One page usually carries most of it. Find it and make it better rather than making more pages.
If you can’t answer those three, you’re producing content, not marketing.
The thing AI can’t do
It can’t have an opinion that costs you something.
Every piece of marketing that has ever worked for a small business contained a claim someone could disagree with. “This is the right way to do it, and the popular way is wrong, and here’s why.” That’s what makes a stranger stop scrolling — not polish, not volume, not optimisation.
A model’s default is to average, and averages don’t take positions. You have to supply the position. AI can help you defend it, structure it, and distribute it. It cannot supply the willingness to be wrong in public.
That’s the whole competitive advantage available to a one-person business, and it’s the one thing that doesn’t get cheaper when production does.
Where the content system fits
The two-week plan above is the manual version of what the Content Engine formalises — a 13-week theme ladder, a day-by-day calendar, 60 pre-written hooks and 12 post templates, so the “what do I write this week” question is already answered.
But the manual version costs nothing and starts tonight. Write down the twenty questions first. Everything else follows from having them.