AI Copywriter Tools and the Scoreboard Problem

Somebody is paying $24.82 for a click on the phrase AI copywriter. That's the cost per click my keyword data returned in July 2026, against about 1,600 searches a month at a difficulty score of 27. Twenty-five dollars a click is serious money, and it tells you the buyers on this query are businesses rather than curious hobbyists.

Here's the short version. An AI copywriter is very good at producing variants and drafts, and it has no way of knowing whether any of them work, because copy is the one kind of writing that gets scored by something outside the document. The scoring happens weeks later, in an analytics dashboard, at a sample size most small sites never reach. Every ranking page for this term compares features. None of them talk about the loop that actually tells you if the copy was any good.

That gap is what this piece is about.

The Only Writing Job With A Scoreboard

Most writing is judged by taste. A blog post is good if it reads well, an email is good if the recipient doesn't get confused. Copy is different, and the difference is the entire discipline.

Copy has a number attached. Clickthrough, signups, purchases, replies. You can be delighted with a headline and be wrong about it, and the traffic will tell you so, eventually, if you get enough of it. That's why copywriters historically got paid well and why the good ones are relentless about testing rather than about adjectives.

A model does not have access to that number. It has access to a large quantity of copy that existed before, which is a different thing, and it's biased toward whatever phrasing was common rather than whatever phrasing converted. Nobody publishes their losing variants, so the training data is a survivorship sample of things people were proud of.

How Long Until You Know It Worked

Here's the arithmetic that made me stop treating "generate 40 headlines" as a productivity feature.

Say your landing page gets 200 visitors a week and converts at three percent. That's six conversions. Six. If a new headline moves you to four percent, you get eight, and the difference between six and eight in a week is indistinguishable from a Tuesday.

Two things fall out of that table. First, the assets with the fastest feedback are the ones where you're buying traffic, which means the loop closes fast only if you're paying for it. Second, on a small site the slower rows never close at all, and you end up choosing copy on judgment, same as always.

I'd argue that's fine. Just be honest that you're picking with taste rather than data, because pretending otherwise is how people talk themselves into rewriting a page every fortnight and calling it optimisation.

What The Model Can't Get From Your Website

The copy brief is the whole job. A model will happily write without one, which is the trap.

The bottom two rows are the ones tools sell. The top six are the ones that decide whether the copy works. That imbalance is not a criticism of the tools, it's just where the line sits, and it explains why the same product feels magical to one person and useless to another. The person who arrives with a filled-in brief gets a fast draft. The person who arrives with "write me a landing page for my SaaS" gets a landing page for a generic SaaS, and it will be smooth, and it will be about nothing.

The Most Expensive Thing On That Results Page

I went through the top ten properly, and the pricing spread is the most interesting thing on it.

The tools cluster low. Alex Birkett's roundup, which sits at position three, lists prices for twelve products and they run from Wordtune at $9.99 a month up to Surfer at $89 for the Essential plan, with Jasper at $24, Copy.AI at $35, and a couple of older names at $57. Loaded 28 July 2026. Jasper's own copywriting page, at position five, publishes no price at all, just a free trial button and a wall of enterprise logos.

Then there's position seven, which is Copy Hackers selling a training course. Not software. $197 a month for three months, or $491 as a one-off, marked down from $591, with a seven-day guarantee. About five hours of lessons.

So the most expensive product on a results page about AI copywriting is a human being teaching strategy, at roughly five times the annual cost of the software. And their own sales page says why. "You still need to figure out what the copy is going to be about. YOU still need the strategy, the plan, the message map." They also write that "airplanes crash when pilots outsource thinking to the machine", which is a bit much, but the underlying point survives the metaphor.

I have no way to verify the outcome they claim for course buyers, so I'm not quoting that part as though it were established.

Twelve Tools, Zero Numbers

Since the roundup is the format that ranks here, it's worth saying exactly what it does and doesn't contain.

Birkett's piece compares its twelve tools on a best-for category, a price, a G2 rating where one exists, and a features list, plus his own experience with several of them. It's a reasonable article, better written than most of its kind, and I'd say the prices alone justify its position.

What it does not contain, and what none of the other ranking pages contain either, is a single measurement of whether copy produced by any of these tools performed. No conversion figures. No before-and-after on a real page. Not one holdout test. The entire category is evaluated on how the output feels to read.

Which is not a scandal, honestly. Running a clean test on twelve tools would take months, need real traffic on each variant, and probably cost more than the affiliate revenue on the article. I understand why nobody does it. But you should read every roundup in this space knowing that "best" means "nicest to use" and not "highest converting", including the ones that put a number in the title. The wider roundup problem across writing tools has the same shape, and it gets worse the more tools a list claims to have tested.

Where It Genuinely Earns Its Keep

I don't want this to read as if the tools are pointless, because they aren't, and I use one most weeks.

Variant generation is the real win. If you've done the brief, know the offer, and have a specific angle, asking for twenty subject lines against that angle takes a minute and two or three of them will be better than what you'd have written cold. Not because the model is a better writer. Because you're a better editor than you are a generator at nine in the morning, and picking from twenty is a different cognitive task than producing one.

The second win is format translation. You wrote a long page, now you need the ad version, the email version and the three-line version. That's mechanical, the substance already exists, and a model does it in seconds without losing the argument. I go through the ordering that works for this in using AI for writing without the failure modes.

Third, and this one is underrated, it kills the blank page. A bad first draft you can react to beats a cursor blinking at you. Whether the draft survives is a separate question and usually the answer is no.

The Free Ones, Briefly

QuillBot has a free AI copywriter page sitting at position four. Loaded the same day, it says it's free to use with no sign-up required and marks itself as Beta. No word cap published anywhere on the page, no price for the premium tier, just an upgrade button.

That's the pattern across most of these free tool pages and I've written about what free actually costs on this category at length, so I won't repeat it here. Short version, the absence of a published cap is not the same as the absence of a cap.

Reading The Results Page Itself

One more thing worth noticing, since it's the sort of signal I now look for before writing anything.

Position two on this query is a Reddit thread titled "Copywriters: What AI tools are you actually paying for". Position ten is a Medium post from a copywriter describing his own workflow. Two of the ten slots are ordinary people talking, on a query where a click is worth twenty-five dollars.

That combination is the single most useful thing a small site can find. When a commercially valuable query has forum posts and personal blogs holding positions in the top ten, it means the well-funded pages have not fully answered the question, and there's a slot available to anyone who writes something genuinely better than a thread. It's not a guarantee of anything. It's just the closest thing to an opening that exists, and it beats every difficulty score I've ever looked at. There's more on how I read a results page in the pillar on what these tools cost and don't fix.

What I'd Do

Write the brief first, by hand, badly. Who's buying, what they're worried about, what the offer is, what the alternative is if they don't buy. Twenty minutes.

Then let a model produce variants against it, pick two, and ship the one you can defend. Don't buy a dedicated copywriting product until you've hit a wall with the general model you already have, because most of these tools are prompts and templates around the same handful of models, and the templates are the cheap part.

And if you can't measure the result, say so out loud rather than calling the change data-driven. Six conversions a week does not settle an argument.