Judge an AI tool on whether it holds your context, whether it does something after generating text, and whether you can get your data out. Tools that only turn a prompt into a paragraph are worth very little, because a general assistant already does that for less.
A week of real work is enough to tell. A demo is not.
The three questions that separate them
Ask these before the pricing conversation.
Does it know your business, or does every session start from nothing. A tool that has to be re-taught your voice, audience and product every time is a prompt box with a subscription.
Does it do anything after writing. Publishing, scheduling, sending, measuring. A tool that hands you text has automated the fastest part of the job and left the slow parts alone.
Can you leave with your work. If your content, contacts and history cannot be exported, the price is not the subscription, it is the switching cost you are accumulating.
Categories worth paying for
Ranked by how reliably they return the money for a small team.
- Transcription and meeting capture, which is cheap and immediately useful
- Anything that reads a large pile of text and tells you the themes: reviews, tickets, survey responses
- Tools that publish and measure, not just generate
- Image and asset generation, if you currently pay for stock or design
- Customer support assistance with a human in front of anything sent
The signs of an expensive wrapper
Some products are a general model with a form on top, priced as though they were more.
Warning signs: output that is noticeably generic no matter what you enter, no way to give it standing context about your business, a feature list that is a list of prompt templates, and pricing per generation rather than per outcome.
The test is simple. Give the same instruction to a general assistant and compare. If the results are similar, you are paying for the interface, and you should decide whether the interface is worth it rather than pretending it is something else.
How to evaluate in a week
Do not evaluate on a sample task. Use it for a week of real work and count what changed.
Day one, give it everything it will accept about your business. Days two to four, use it for whatever you would have done anyway. Day five, compare the output to what you would have produced and note how much editing it needed.
The number that matters is how much of the output you shipped without substantial rewriting. Below half, and the tool is generating work rather than saving it.
The cost people forget
Subscription is the small cost. The large ones are the time to learn it, the time to feed it context, and the risk of building a process around something that may change or disappear.
Small companies commonly end up with six AI subscriptions that overlap, none of which know about each other, which recreates the tool sprawl the tools were meant to reduce.
Before adding one, check whether something you already pay for now does it.
When not to buy anything
If your constraint is knowing what to do rather than producing it, a generation tool makes the problem worse by producing more of the wrong thing faster.
If you have no measurement, adding output means adding things you cannot evaluate. Fix that first, or the tool will make you busier and no wiser.
Common questions
- How do I choose an AI tool?
- Ask three things: does it hold context about your business, does it do anything after generating text, and can you export your data. A tool that only turns a prompt into a paragraph is worth little, because a general assistant already does that for less.
- How can I tell if an AI tool is just a wrapper?
- Give the same instruction to a general assistant and compare. If the results are similar you are paying for the interface, which may still be worth it, but decide that knowingly.
- How should I evaluate an AI tool?
- Use it for a week of real work, not a sample task. The number that matters is how much output you shipped without substantial rewriting. Below half and the tool is generating work rather than saving it.
Keep reading