TL;DR: DIY marketing with AI is a real option, and price is not what should decide it. The deciding factor is whether the work sits inside AI’s current capability or outside it. In a field experiment with 758 BCG consultants, AI helped inside that boundary and made people 19 points less accurate outside it. Most DIY decisions fail on that line, not on budget.

First, the conflict of interest

I run an SEO and GEO agency. If you read this and decide to hire someone, there is a version of that where you hire us. Weigh everything below accordingly.

Here is the counterweight, and it is the only reason I think I can write this honestly. We turn work away fairly often, and not out of virtue. A client who should have kept the work in house churns by month four, feels badly served, and costs us more than the retainer ever brought in. So my incentive is less lopsided than it looks. I would rather talk you out of hiring us in week one than lose you in month five.

Every study I cite below is linked. Go and check them.

The budget question is not the real question

This conversation almost always starts the same way. A small business owner, founder, or solo marketer has an agency quote sitting in their inbox, ChatGPT open in another tab, and wants to know whether AI marketing tools for small business can do the job instead. Often, they can handle specific marketing tasks well—first drafts, keyword list expansion, meta descriptions, reformatting, summarizing data, and some AI-powered features tied to content, design, and marketing automation—but not the whole marketing strategy, and not without review. Reasonable question. Wrong frame.

The better question is which specific jobs in your marketing sit inside what AI currently does well. AI marketing tools can automate workflows, optimize marketing tasks, and improve personalization for small businesses, but those gains only help when the work stays inside AI’s reliable range, and when someone on your side can tell when a job has quietly crossed the line. That is the real DIY-versus-agency decision this piece looks at, along with cost, capability, risk, and when outside help is worth paying for.

That framing is not mine. Researchers from Harvard Business School, Wharton and MIT ran a pre-registered field experiment with 758 BCG consultants, and after peer review it was published in Organization Science in March 2026. They named the boundary the jagged technological frontier. On tasks inside it, consultants using GPT-4 completed 12.2% more tasks, worked 25.1% faster, and produced output rated more than 40% higher in quality. On a task deliberately chosen to sit outside it, consultants using AI were 19 percentage points less likely to reach the correct answer than consultants with no AI at all.

Sit with that second number for a moment. Not “no better.” Worse. Worse than not using it.

For marketing, the danger is the failure mode rather than the failure rate. The same AI-powered tools that automate repetitive tasks still need human oversight when judgment is involved. AI does not stop and tell you it has wandered past the edge of what it knows. It writes something fluent, confident and wrong, and the only thing standing between that and your website, campaign, or broader marketing efforts is a person who knows enough to catch it.

You are a bad judge of your own speed

This next one bothers me, because I have caught myself doing it.

In July 2025, METR ran a randomized controlled trial with 16 experienced open-source developers across 246 real tasks in repositories they knew well. The developers predicted AI would make them 24% faster. Measured, it made them 19% slower. Afterwards, having lived through the slowdown, they still believed AI had sped them up by about 20%.

The caveats are real and I want them on the record. Sixteen people is a small sample. It was software, not marketing. And METR posted an update in February 2026 saying the 2025 result is historical, that later data showed some evidence of speedup, and that selection effects made the newer numbers hard to interpret, which is why they are redesigning the experiment. I would not use this study to claim AI slows anyone down today.

What survives is narrower and still useful. How fast you feel is not evidence of how fast you are. And nearly every DIY business case I get shown is built on how fast somebody felt. “I wrote three posts on Sunday.” Fine. How long was the editing, how many did you bin, and what did those three posts do in ninety days?

What DIY actually costs

Software is the knowable part. I checked all three major SEO platforms’ pricing pages myself in August rather than trusting my own older post, and wrote it up in Ahrefs vs Semrush vs Moz. Moz Pro Standard is $99 a month, Ahrefs Lite is $129, and Semrush’s SEO plan is $139. If you want AI visibility tracking on top, Moz bundles 50 tracked prompts into that $99 plan, Semrush’s AI search coverage starts at $199, and Ahrefs sells Brand Radar separately from $199.

So the floor is around $99 a month, and a single person who wants proper AI visibility tracking lands somewhere in the low hundreds. That is not the expensive part.

The expensive part is you. I am not going to hand you an industry-average retainer figure, because most of the numbers in circulation are agency marketing, and that includes mine. Run your own instead. Take the hours the work honestly needs each month, multiply by what an hour of your time is worth when it is spent on the thing only you can do, then add the software. Put that next to the quote in your inbox.

Founders who run that calculation properly tend to get surprised in one direction or the other, and both answers are legitimate. The thing that is not legitimate is skipping the calculation and calling DIY free because no invoice turns up.

The playbook you learned may be pointed at the wrong marketing strategy target

There is a second problem with DIY that has nothing to do with effort, and I think it catches more people than the first one.

The Pew Research Center tracked the browsing behaviour of 900 US adults across 68,879 Google searches in March 2025. When an AI summary appeared at the top of the page, people clicked a traditional result in 8% of visits, against 15% when no summary appeared. Clicks on a link inside the summary itself ran at about 1%. Pew is careful to call this an association rather than proof of cause, the data is Google-specific, and it is over a year old now. The direction is still hard to argue with.

If you learned marketing in the era where ranking produced clicks and clicks produced leads, the machinery you are pointing your weekends at may be measuring something that has weakened underneath you. Getting cited inside the answer is now its own job with its own methods. DIY rarely fails on execution. It fails because someone is doing last year’s work very diligently.

Four questions that settle it

Skip the pros and cons lists. These four do the actual work.

Which side of the frontier is this task on?

Inside, in my experience: first drafts, using ai writing to create content for blog posts, speeding up the content creation process, expanding a keyword list, meta descriptions, reformatting, summarising a Search Console export, supporting content marketing with routine content creation, shaping a content strategy, and getting unstuck on angles while keeping a consistent brand voice. Outside: deciding which page to kill, judging whether a link is worth having, and reading a ranking drop where five explanations all look plausible. The second list is where money is won and lost, and it is precisely where the research says unaided humans do better.

Who on your side catches a confident wrong answer?

If nobody on your team can look at a fluent, assured recommendation and say “no, that is wrong, and here is why,” then AI does not multiply your output, even when it can analyze customer behavior, consumer behavior, individual customer behavior, customer data, and complex data across many data points to surface actionable insights, consumer insights, customer feedback, the customer journey, customer relationships, and patterns among potential customers in real time for marketing teams. It multiplies your error rate, quietly, at volume. That matters most when you use it to engage customers through personalization, better lead scoring, or audience segmentation based on real-time signals, because AI tools can automate audience segmentation based on real-time behavior and improve segmentation accuracy, but the output still needs validation. This one question decides more DIY outcomes than budget does.

Are you buying tools for content creation or building a system?

MIT’s Project NANDA reported in July 2025 that roughly 95% of enterprise generative AI pilots produced no measurable P&L impact, and blamed a learning gap: the systems did not retain feedback, adapt to context, or improve with use. That report has not been peer reviewed and its methodology has taken fair criticism, so hold the exact number loosely. The pattern held in my own agency, where I once counted eleven AI subscriptions and could not name a single thing in the business that had changed because of them. Disconnected subscriptions do not compound when they sit outside existing marketing workflows and existing systems. An AI layer tied into your tech stack can use Zapier to connect business apps, support internal tools, and automate data gathering or cleansing across the marketing workflow.

What does six months of nothing cost you?

If half a year of flat numbers is survivable, do it yourself, learn it properly, and keep the money. If half a year of flat numbers means a missed raise, a missed hire, or a competitor taking the category, then you are not buying labour when you hire someone. You are buying a deadline. Price it that way.

When I tell people not to hire us

One site, a niche where nobody serious is competing, and a founder willing to spend real hours. Do it yourself. Buy Moz, work through Google’s own SEO starter guide, publish consistently, and come back in a year.

Pre-product-market-fit is the other one. If you are still changing who you sell to every quarter, a retainer is an expensive way to produce content about a positioning you are going to abandon.

And if what you need is the boring basics done once, that is a project, not a retainer. Say so, and pay for a project.

When hiring is worth it for a small business

A category with entrenched competitors, where the gap is authority rather than effort. Link acquisition, which is relationship work and does not compress no matter how good your prompts get. An in-house marketer already at capacity, where the constraint is hours rather than knowledge. Or a deadline attached to revenue.

If you do go looking, the thing I would insist on is a baseline number before you sign anything. If an agency cannot tell you where you stand today across a defined set of prompts and queries, they will construct a flattering starting point later. I went into how to vet that in AI SEO agency: what they do differently.

Where I land

The real answer for most businesses is not a side. It is a split. Use AI for the inside-the-frontier volume, which is genuinely a large share of the work. Keep a human on the outside-the-frontier calls. Buy out the pieces that need relationships or time you do not have.

Before you decide either way, write down the single number you would need to see six months from now for the decision to have been the right one. Traffic is usually the wrong number. Pipeline is usually the right one. If you cannot name the number, no agency and no model is going to fix that for you, and that is the most useful thing I can tell you today.

FAQ