There are more AI marketing tools than anyone could evaluate, and a new one is recommended to your CEO on LinkedIn every week. Many are a general-purpose AI model with a narrower interface and a subscription attached.
So the sensible approach is simple. Decide which jobs you want help with, start with the one tool that covers most of them, and add others only when you can name the job the first one does badly.
Start with one general assistant
A general assistant is a chat-based AI you can give instructions and documents to. ChatGPT, Claude and Gemini are the well-known ones. Any of them will draft, summarise, restructure, critique, brainstorm and explain. That covers most of what the specialist tools promise.
Pick one. If your company already pays for a workplace suite that includes an assistant, use that, because it is likely already approved for company data. Then spend a month building habits before you buy anything else:
- Give context every time. Who the audience is, what the piece is for, what you already know, what good looks like. Our guide to briefing AI and humans has a template.
- Keep a reference document. Your tone of voice, your products, your customers, phrases you never use. Paste or attach it at the start of each piece of work.
- Ask it to critique before it creates. “What is weak about this draft?” is often more useful than “write me a draft”.
- Check every fact. Every one.
- Save prompts that work. A plain document of your ten best is worth more than a new subscription.
A second tool will not fix a problem caused by a poor brief to the first one.
The stack by job
| Job | Good for | Where it fails | The human check |
|---|---|---|---|
| General assistant | Drafting, summarising, restructuring, thinking aloud | Confident errors, bland default voice | Verify facts, rewrite in your voice |
| Research | Fast overviews with sources, competitor scans | Misread or weak sources, stale information | Open the sources and read them |
| Writing and editing | Tightening, grammar, consistency, repurposing | Flattens personality, changes meaning | Read it aloud, compare with the original |
| Images and design | Layouts, resizing, concept visuals | Generic look, odd details, rights questions | Brand check, zoom in, confirm usage rights |
| Meeting notes | Transcripts, summaries, action lists | Wrong names, missed nuance, consent | Correct before sharing, tell attendees |
| SEO and analytics | Spotting patterns, clustering queries, drafting titles | Invented data, advice with no context | Check against Search Console and analytics |
| Automation | Moving information between systems | Silent failures, errors repeated at scale | Approval step before anything is sent |
Research
Research tools search the web and write an answer with citations. Perplexity is the best-known dedicated example, and the general assistants can do the same. They are good for getting oriented in an unfamiliar subject.
They fail by citing a source that does not say what the summary claims, or by treating a weak source as an authority. Open every link you intend to rely on. Never put a statistic in your own content because an AI told you a study said so. Find the study.
Writing and editing
Tools built into your document editor, and dedicated writing assistants, are useful for cutting a long draft down, fixing grammar, and turning one piece into several formats.
Their weakness is that they pull everything towards the same safe, smooth register, which is why so many brands now sound alike. They will also quietly alter meaning while “improving” a sentence. Use them on a draft that already has your opinions in it, and reread the result against the original.
Images and design
Two different things live here. Design tools such as Canva use AI to speed up layout, resizing and background removal, and that is low-risk and helpful. Image generators create pictures from a text description.
Generated images are fine for internal concepts and mood boards. For public use, be careful: they tend towards a recognisable look, they get details wrong (hands, text, logos, machinery), and the legal position on ownership and on what they were trained on is not settled. Read the tool’s commercial terms, and do not generate images of real people.
Meeting notes
A notetaker records a call and produces a transcript, a summary and action points. It lets you pay attention in the meeting.
It will mishear names and figures, and it cannot tell a firm decision from a passing remark. Correct the summary before circulating it. Tell everyone on the call that it is being recorded, every time, and check that your clients’ contracts do not forbid it.
SEO and analytics
AI is good at pattern-finding in data you supply: grouping a long list of search queries into themes, drafting title and description options, explaining a chart, suggesting what to look at next.
The risk is using it as a source of data. A general assistant asked for keyword search volumes may simply make numbers up. Real figures come from real tools: Google Search Console, Google Analytics, and dedicated SEO software. Export the data, hand it to the assistant to interpret, and check its conclusions against the source.
Automation
Automation tools connect your systems so that an event in one triggers an action in another, such as a form submission creating a CRM record.
This is where mistakes multiply. A person makes an error once. An automation makes it every time until somebody notices. Put a human approval step in front of anything a customer will receive. Have failures alert you, and review what it has been doing each month.
How to trial a tool for a fortnight
Run a free trial as a test.
- Pick one real, recurring task. Something you do at least weekly, such as the monthly report, turning a webinar into posts, or first-draft product descriptions.
- Write down your baseline. How long it takes you now and how good the result is.
- Define success in advance. For example: “Cuts this from three hours to one, and the output needs no more than twenty minutes of editing.”
- Use it on that task every time for two weeks. Do not wander through the feature list.
- Count the checking time. The honest figure is total time to a finished, correct result, including fixing what the tool got wrong.
- Compare with your general assistant. Run the same task there with a decent prompt. If the result is close, you do not need the new tool.
- Decide on day fourteen. Keep or cancel. Set a calendar reminder before the trial converts to a paid plan.
Questions to ask before company data goes in
Before pasting anything about customers or anything confidential into a tool, find out:
- Is what you enter used to train the provider’s models, and can that be switched off?
- Where is the data stored, and for how long?
- Who else in your organisation can see your conversations?
- Does the account sit under a company login you control, or someone’s personal email?
- Does your AI policy cover it?
If you cannot get answers, keep sensitive information out of that tool.
When to stop buying
A reasonable stack for a one-person team is short: one general assistant, a design tool, a notetaker if you spend a lot of time in meetings, and the analytics you already have. Review the list each quarter and cancel whatever you have not opened in a month.
The tool is the easy bit. The quality of what comes out depends on the brief going in and the person checking the result, which is the principle our own work is organised around. For the tasks where the saving is real, see where AI saves a one-person team time.