AI writing tools: what people actually complain about

You've been told you should be publishing content, and there's a row of AI writing tools happy to sell you a subscription for it. Before you pick one, it's worth knowing what the people already paying for them are unhappy about. This post goes through the complaints that come up again and again, what's actually behind each one, and how to test for them in a trial — so you buy with your eyes open or don't buy at all.

First, who's doing the complaining

With Jasper — probably the best-known dedicated AI writing tool — 93.5% of reviewers come from small businesses of 50 staff or fewer. That matters. These aren't enterprise content teams with a strategist and two editors. They're people roughly in your position: one or two folk trying to keep a website looking alive alongside the actual work. So the complaints are relevant to you in a way that enterprise software reviews usually aren't.

There are five that recur. Four are about the writing tools themselves. One is about the SEO tools people bolt on alongside them.

Complaint 1: the output is generic and still needs editing

This is the big one. The most common criticism of Jasper is that without detailed prompting it defaults to safe, predictable phrasing — a competent first draft with no point of view. Reviewers describe repetitive output that reads fine and says nothing.

It's worth understanding why, because the fix follows from it. These tools work by producing the most likely next words. Ask for "a blog post about boiler servicing" and you get the average of everything ever written about boiler servicing. Which is, by definition, unremarkable. Nobody's average article mentions that you charge £95 including the gas safety check, or that the most common reason a callout turns into a full swap is a knackered diverter valve, or that you won't touch a certain brand of combi any more.

So the honest way to think about it: it's a drafting tool, not a writing tool. You supply the specifics, the opinions and the prices; it supplies structure and speed. If you were hoping to press a button and get something publishable, the complaints are telling you the truth and you should believe them.

The practical consequence is that the time saving is smaller than advertised. Budget for editing. If a post takes you three hours from scratch and ninety minutes with a draft in front of you, that's a real saving — but it's a saving of ninety minutes, not three hours, and the tool has to be cheap enough to justify it.

Complaint 2: the price, relative to what you get

"Expensive" turns up constantly in negative reviews of Jasper. Its entry self-serve tier is $69 a month, which sits near the top of the category. That's the complaint in one line: people don't object to paying, they object to paying premium rates for drafts they then have to rewrite.

Complaint 3: "it's a wrapper with a good UI"

A vocal minority on Reddit dismiss Jasper as a wrapper — a nice interface sitting on top of the same underlying models anyone can reach for free. Several say they stopped paying once general-purpose chat tools got good enough to do the same job.

This is partly fair and partly lazy. It's fair in that the raw text generation really does come from the same family of models. It's lazy in that the wrapper is sometimes the point — templates, a saved brand voice, somewhere for outputs to live, seats for two people, an interface your part-time admin can use without learning to prompt. That scaffolding has value if someone is producing content most days.

But if you're one person writing occasionally, the criticism lands. There's a cheap test for it, and you should run it before paying anything: spend a fortnight doing the job in a free general chat tool. Notice precisely what annoys you — losing your brand guidelines every session, re-pasting the same instructions, no way to hand it to someone else. Then look at whether the paid tool fixes those specific annoyances. If it doesn't, you're buying a nicer screen.

Complaint 4: the reviews don't agree with each other

Jasper's Trustpilot rating is 3.3 out of 5 — noticeably worse than its scores on G2 and Capterra. That gap confuses people, and it shouldn't. The sites collect reviews from different populations. Software marketplaces tend to catch users mid-use, often prompted. Trustpilot tends to catch people who went looking for somewhere to complain — usually about billing, renewals or cancelling.

The useful move isn't averaging the scores. It's reading the one- and two-star reviews and sorting them into two piles: complaints about the product and complaints about the commercial relationship. Product complaints you can test in a trial. Billing complaints you can't — and those are the ones that predict how you'll feel in month eleven. We found exactly the same pattern when we looked at the complaints about Podium, Birdeye and NiceJob: the software criticism was mild, the contract criticism was not.

Complaint 5: SEO tools that talk you into writing for the algorithm

Separate category, same buying decision. Surfer-style SEO content tools score your draft against competing pages and tell you which terms to add and how to structure headings. The recurring criticism on Reddit is that following that guidance too literally produces over-optimised, heading-stuffed content that reads like it was assembled for a crawler rather than a person. "Write for users, not search engines" is the pushback, and it's said often enough that it's clearly hitting something real.

Here's the nuance the complaints usually miss: used as a guide rather than a rulebook, the criticism softens considerably. The term list is genuinely useful as a prompt for things you forgot to cover — if every competing page on emergency callouts mentions parking access and you didn't, that's a gap worth filling. The trouble starts when the score becomes the goal and you start wedging phrases into subheadings to move a number from 74 to 89.

A simple rule: if you can't say the heading out loud without wincing, delete it. No score is worth a page your customer bounces off.

The complaints, and how to test each one

Where these tools aren't the right fit

Some honest cases where you should walk away from the whole category.

• You publish rarely. Four posts a year doesn't justify a subscription. Write them yourself, or pay a human who understands your trade.
• Your claims are regulated. Clinics, aesthetics, anything financial. A tool that confidently produces plausible-sounding claims is a liability, not an asset, when someone else has to stand behind the wording.
• The content depends on first-hand knowledge. Case studies, pricing pages, "what we found when we opened the wall". No model knows what happened on that job. You'll spend longer correcting the invented version than writing the real one.
• Your problem isn't content at all. This is the common one. Plenty of businesses buy a writing tool when the actual leak is elsewhere — enquiries nobody answers, or no outbound activity at all. If the phone rings while you're under a sink, more blog posts won't help; the arithmetic on that is in what missed calls really cost a UK trades business. If the site simply gets no traffic yet, you may need a way to reach businesses directly before content has anything to work with — contact-form outreach is one route.

It's also worth noting that the backlash against AI writing tools is mild compared with the backlash against AI tools that contact people on your behalf. Bad copy wastes your afternoon. Bad automated outreach damages your reputation and your email deliverability — we went through why in the AI SDR trust problem. Different risk profile, different level of caution required.

The question that predicts whether you'll still be paying in a year

Not "is the output any good?" Everyone's output is roughly the same standard, because it's roughly the same models underneath. The question is: how many hours a month does this genuinely remove, and would I pay that money for those hours in cash?

Work it out in your own numbers. If a tool saves you ninety minutes a post and you write two a month, that's three hours. Three hours of your time is worth whatever you'd bill for it, or whatever you'd pay someone else. If that figure comfortably clears £54 a month, the subscription is fine and the complaints about generic output are just a description of the work you still have to do. If it doesn't clear it, no amount of interface polish will fix the sums — and the reviewers grumbling about price on Trustpilot have already done the arithmetic for you.