AI Writing Tools, and What They Actually Cost

Eight of the nine organic results my checker returned for AI writing tools came from somebody with a product to move. I pulled that top page in July 2026 and counted them off. Grammarly, QuillBot, Ahrefs, DeepAI, TinyWow, Scribbr, plus two roundups. The ninth is a Reddit thread, and it's the only result where somebody is talking rather than selling.

So this is the other kind of page. Less enthusiasm, more arithmetic.

Short answer, for anyone skimming. AI writing tools do four separate jobs that get marketed as one, and the price gap between doing those jobs through a subscription versus through raw API tokens is larger than most people realise. A 2,000-word draft costs somewhere around three cents in tokens on a cheap model, or twelve dollars a month on a subscription plan you may only half use. I'll show both sets of numbers below. And the more important thing, which I learned slowly and expensively, is that none of these tools touch the part that decides whether anyone ever reads the finished piece.

Who Owns This Search Result, And Why It Matters

I run a keyword pipeline against a paid keyword-data API, plus a live SERP checker, mostly so I can stop guessing about this stuff. In July 2026 it reported roughly 60,500 monthly searches for this phrase at a difficulty score of 26, which sounds approachable until you look at who's already there.

Here's what the results page actually contained:

Five of those nine are free tool pages published by the tool's own vendor. That's not a conspiracy, it's just what happens when a query has commercial intent and the vendors have domain authority. But it does mean that if you're reading roundups to decide what to buy, you are almost entirely reading marketing, and the one place where somebody says "actually that one was disappointing" is a forum post.

The Reddit result is the useful signal, incidentally. When a results page has a forum or a Medium post in it, that's a slot a smaller site can genuinely take. It's the single most reliable thing my scans surface, and I'll come back to it.

Four Jobs, Sold As One Product

Nobody sits down to "use an AI writing tool". They sit down with a specific problem, and the problem determines which category is relevant.

Drafting. Getting words on a page from a brief or an outline. This is what a general model does, and honestly it's the part that's least differentiated between products, because most of the writing tools are wrapping the same handful of models with a different interface on top. If you already pay for a chat interface, you already have a drafting tool. There's a whole family of queries around AI writing generators and chatgpt as a writing tool that all land in this category.

Editing. Grammar, consistency, tone. Grammarly is the incumbent and it's very good at the narrow thing it does. This is genuinely different from drafting and worth paying for separately if you write a lot.

Rewriting. Taking existing text and producing a different version of it. Paraphrasers, humanisers, spinners. It's the most crowded corner of the market and, I'd argue, the least understood by the people buying it. I wrote a whole piece on what an AI re writer actually changes because the search results for it contain zero explanatory articles, only tools.

Research and structure. Finding out what to write about, what's already ranking, what's missing. This barely overlaps with the others. It's closer to using AI for SEO than to writing, and it's the part that most affects whether the finished piece does anything.

The reason the categories matter is that a subscription buys you one of them. Buying three subscriptions to cover three jobs is how people end up spending sixty dollars a month on writing three posts.

What A Draft Costs In Actual Money

This is the section the roundups never write, presumably because it makes the subscription look worse.

Model pricing is public. Anthropic's pricing page, which I loaded on 28 July 2026, lists Claude Sonnet 5 at an introductory $2 per million input tokens and $10 per million output through 31 August 2026, going to $3 and $15 after that. Claude Haiku 4.5 sits at $1 in and $5 out. Claude Opus 5 is $5 and $25. The same page gives a rough conversion of one token to about 0.75 words in English, which is close enough for planning.

So a 2,000-word article is roughly 2,700 output tokens. Input depends entirely on how much research you're stuffing in front of the model. Here's the spread, using a bare prompt versus a research-heavy one at about 40,000 input tokens:

Twenty-seven cents. For a fully researched 2,000-word draft on the most expensive model in that table, with a large pile of source material fed in first.

I want to be careful not to oversell that. Those are the raw inference costs and they exclude the failed attempts, the reruns, the version you throw away because the angle was wrong, and your own time reading it. In practice I'll burn three or four drafts before one survives, so multiply by four and you're still under a dollar and change. The point isn't that writing is now free. It's that the marginal cost of the drafting layer has collapsed to something that no longer belongs in the same conversation as a monthly subscription.

The Subscription Math, Which Cuts Both Ways

Now the other side, because I'm not arguing everyone should be hitting an API.

I pulled these from the vendors' own pricing pages on 28 July 2026. Grammarly runs a free tier at $0 with 100 AI prompts a month, Pro at $12 per month with 2,000 prompts per member, and an Enterprise tier with unlimited prompts and no published price. QuillBot's Premium page showed $4.17 per month billed annually, with a free tier that caps the paraphraser at 125 words, the humaniser at 125 words and six uses a day, the summariser at 1,200 words, and the plagiarism checker at 25,000 words a month.

I tried to add a couple more vendors to that list and one of the pricing pages returned a 403 to me twice, so it isn't here. I'd rather leave a hole than quote a price I didn't personally load.

The honest read on those numbers is that they're cheap and the free tiers are less crippled than you'd expect. QuillBot at fifty dollars a year is not the thing that's holding anybody's content back. Grammarly Pro at twelve is a rounding error against the value of not shipping a typo in a client deliverable.

Where subscriptions stop making sense is when you're paying for three of them because each covers a different one of the four jobs above, and you're using maybe fifteen percent of each. That's the trap. Not the price of any single tool, but the stack.

If you write one post a week, one subscription plus a chat interface covers everything. If you're running a content pipeline at volume, the token math wins by a distance and it isn't close. There's a middle zone in between where the answer is genuinely unclear and I don't have a clean rule for it, which is annoying but true.

The Free Tier Ceiling, Measured

Everybody says "the free tier is enough to start" and nobody converts that into a number, so I did.

Take a single 2,000-word article and run it through what the free plans permit, using the caps published on those same pricing pages.

The paraphraser row is the one worth sitting with. Sixteen copy-pastes to rewrite one article. That's not a limit you hit and notice, it's a limit that quietly makes the free tier unusable for the exact job people signed up for, and it's clearly deliberate.

The plagiarism checker at 25,000 words a month is the most generous of the lot and probably the least used. And Grammarly's 100 prompts is genuinely fine if you write once a week, which describes most people who think they need a paid plan.

My read, and I hold it loosely, is that the free tiers are calibrated to be sufficient for someone evaluating the product and insufficient for someone using it. Which is a reasonable thing for a business to do. It just means "try the free version first" is advice that expires quickly, and knowing exactly where it expires saves you from paying a month early or struggling a month late.

What Google Says About This, In Its Own Words

Everybody has an opinion about whether AI-written content ranks. Very few of the pages arguing about it quote the actual documentation, so here it is.

Here's Google's helpful content guidance, loaded 28 July 2026. "If you use automation, including AI-generation, to produce content for the primary purpose of manipulating search rankings, that's a violation of our spam policies." The same page asks whether "the use of automation, including AI-generation, self-evident to visitors through disclosures or in other ways".

The spam policies page defines the actual offence. "Scaled content abuse is when many pages are generated for the primary purpose of manipulating search rankings and not helping users." It lists as an example "Using generative AI tools or other similar tools to generate many pages without adding value for users".

Read those carefully and the operative words aren't "AI". They're "many pages", "without adding value", and "primary purpose". The policy is about industrial-scale thin output, not about whether a model helped you write a paragraph. Which is roughly what everyone sensible already assumed, but it's nice to have the sentence rather than the vibe.

There's a related doc on AI features that's worth quoting too, because a whole cottage industry has grown up selling optimisation for AI Overviews. Google's page says "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary" and, more bluntly, "You don't need to create new machine readable files, AI text files, or markup to appear in these features. There's also no special schema.org structured data that you need to add."

I have no idea how long that stays true. Right now it's the documented position.

The Detector Question

Detection is its own small industry and it deserves a paragraph of honesty.

Originality.ai's pricing page, loaded the same day, lists a Pro plan at $14.95 a month for 2,000 credits, or $12.95 a month billed annually. One credit covers 100 words of AI detection, or 100 words of combined AI and plagiarism checking costs two. So a 2,000-word article is twenty credits, which works out at about fifteen cents a scan on the monthly plan.

Fifteen cents is nothing. The problem isn't the price.

The problem is that a scan tells you how a sentence-level statistical pattern looks, not whether the writing is good or true. My own first drafts scan badly even when every number in them came from a source I loaded myself that morning, because the thing being measured is rhythm and predictability rather than accuracy. Fixing the score means writing with more variance, more hedging, more specific detail, and shorter sentences next to much longer ones. Which happens to also make the writing better, so I've stopped resenting it. But treating a detector score as a quality gate is a category error, and I've watched people rewrite genuinely good paragraphs into worse ones chasing a number.

The Part No Tool On That List Does

Here's the receipt that changed how I think about all of this.

I had a site sitting at an average position of 47 in Search Console for a long stretch. Google defines that metric precisely, and the wording matters. It's "the topmost position occupied by a link to your property or page in search results, averaged across all queries in which your property appeared", and a link "must get an impression for its position to be recorded". So position 47 doesn't mean nobody liked it. It means the pages were indexed, they were being served, and they were landing somewhere around page five where the effective audience is zero.

I responded to that the way most people do, which is by improving the content. Better drafts, tighter structure, more research. It was comfortable work and it changed nothing, because the content was never the bottleneck.

The first sale on that site was $29.99 and it landed roughly two months after launch. Across everything I've shipped, all products, all time, revenue is somewhere around ten thousand dollars. Lifetime, not monthly, and I say that explicitly because the number occasionally gets quoted back at me as though it were a monthly figure. Seven digital toolkits priced between $29 and $49 sit inside that total.

None of the writing tools I've listed on this page would have moved any of those numbers. They make the drafting faster and the editing cleaner. Distribution is a completely different problem and it stays unsolved by anything you can buy for twelve dollars a month.

How I Actually Choose

Pick the job first, then the tool. It sounds obvious written down and almost nobody does it.

If you write occasionally and want fewer mistakes, an editing subscription is the right purchase and the drafting tools are noise. If you write at volume, go straight to a model with an API and stop paying per seat for a wrapper. If you're rewriting existing material, understand what you're actually doing to it before you commit, because rewriting tools change the surface and leave the substance alone, and search engines have gotten reasonably good at noticing.

And if the goal is ranking rather than writing, most of the useful work sits earlier in the process than any of these tools operate. Choosing the right thing to write about beats writing the wrong thing beautifully. That's what the research side is for.

Questions That Come Up Every Time

Do AI writing tools hurt your rankings? Not by themselves, per the policy language quoted above. What gets penalised is scaled thin output with no added value. A well-researched piece where a model helped with the prose is not what "scaled content abuse" describes.

Which one is best? Depends entirely on which of the four jobs you're doing, and I'd be suspicious of anyone who answers that question without asking which one first. The roundups answer it because ranking for the question requires answering it.

Are the free tiers usable? More than you'd expect. QuillBot's free paraphraser caps at 125 words per pass, which is annoying but workable. Grammarly's free tier gives you 100 AI prompts a month. For low volume, free is genuinely fine.

Do I need a separate tool for copywriting or long-form fiction? Usually no. The specialised products are mostly prompt scaffolding around a general model. The specialisation is in the prompts, and prompts are free.

How much should I budget? If you're starting, zero. Free tiers plus a chat interface will carry you further than the pricing pages want you to believe. Move to paid when a specific limit is actually blocking you, not before.

What I'd Do With A Blank Page

Start with one editing subscription and whatever general model you already have access to. Don't buy a drafting tool. Don't buy a rewriter at all until you can articulate what you want it to change.

Spend the money you didn't spend on subscriptions on the research layer instead, because the difference between an article that ranks and one that doesn't is decided before the first sentence gets written. A keyword-data subscription and a SERP checker cost less than the three-tool writing stack and they answer a more important question.

Then, and this is the part I keep having to relearn, decide how anyone is going to find the thing before you write it. Position 47 taught me that lesson and it took two months and a single $29.99 sale to make it stick.