Sudowrite: reviews and analysis

TLDR: Sudowrite is the reference point for this whole segment, to the extent that half its rivals market themselves as alternatives to it. Its answer to the forgetting problem is not a filing cabinet of character cards: it is an engine that decides which slice of your story bible the model gets to see on each generation, plus chapter chaining and shared characters across books in a series. The part to inspect closely is the bill. The figure on the pricing page is not the monthly one, and the credits a novel burns can vary tenfold depending on how you work.

What Sudowrite is and how it works

Sudowrite came out of closed beta on 19 November 2021 with a three million dollar round raised almost entirely from individual investors, as TechCrunch reported at the time. That piece described the tool as built on GPT-3, with three to four hundred subscribers paying around twenty dollars a month. Five years on the premise is unchanged, an editor where the AI lives inside the text, but the machinery underneath looks nothing like it did.

The central piece is called Story Bible and it is a chain rather than a form. It starts with the Braindump, which you write by hand and cannot generate, and from the Braindump plus genre comes the synopsis. From the synopsis come characters and worldbuilding. From all of that comes the outline, and from the outline come scenes and draft. Each link feeds the next and the documentation states which depends on which, so changing the genre at the top tells you what shifts below. One warning for anyone arriving via an old comparison: the beats you used to generate chapter prose from have been phased out, replaced by scenes and draft.

What actually answers this segment’s question is the saliency engine, and it deserves unpacking because almost nobody explains it properly. There is no button for it. When you ask for prose, the system does a first pass over everything in your Story Bible and works out which fragments matter for that specific task before handing them to the model. The example in the documentation is a sharp one: if your shifter is in wolf form, the AI will not bring up the pleated khakis they wear to the office. Translated, that is selective retrieval over your encyclopedia rather than dumping the whole thing.

And it comes with the control that mystery writing demands. Every character card and every individual trait can be hidden from the AI with a toggle. You can hide a character who does not appear until halfway through, and you can hide the murderer’s motive trait so the engine never rules it relevant and spoils the ending in a second-act scene.

Two more mechanisms handle long work, and their numbers repay a careful read. Chapter continuity chains documents by telling the tool which one precedes which, and with that the AI reads up to twenty thousand words of the current document plus up to twenty thousand words in total spread across the linked chapters, reaching as far as twenty-five documents back. That total matters: it is not twenty thousand per chapter, it is twenty thousand for all of them together, so on a long novel the distant past arrives trimmed. Series support shares characters, worldbuilding and outline across the books inside one folder, though only within the same level, since a subfolder does not share upwards.

The writing tools themselves are the familiar set and still good. Write generates between fifty and two thousand words across three modes, automatic, guided with your own instruction, and a tone shift with registers like ominous, fantastical or romantic. Describe produces sense-by-sense description, metaphor included. Expand stretches a short passage without exceeding a thousand words per pass. Brainstorm fires off ideas, Canvas lays the outline out as cards, and First Draft writes up to three thousand opening words, but only in a completely empty document.

The in-house model is called Muse and it is the default in Write and in draft. The company says it trained the model measuring clichés in order to strip them out systematically, and states flatly that it carries virtually no content filter, so it writes violence and adult themes other models refuse. Worth knowing before paying, in either direction. Alongside Muse sits a long catalogue of third-party models, and here the documentation is transparent to the point of identifying its own Basic tier as GPT-4o Mini.

What it is like day to day

The picture from reviews with real volume is good and uneven at once. Its iPhone app holds 4.5 out of 5 across 356 ratings, the largest opinion base found anywhere in this segment, and Capterra shows 4.5 out of 5 from only four reviews, far too small a sample to mean anything by itself but useful for what those reviews say.

What they say is this. A review signed by Jamie Biggs at The Write Practice, dated 29 October 2025, reports that the series folder holds story details across projects and that the more of her own material she uploaded, the better the tool grasped her tone, pacing and dialogue rhythm. It is a favourable review, and worth saying that it declares affiliate links and that the author was given access, so it reads with that discount. It still leaves two useful warnings: opening Sudowrite for the first time is intimidating and she found herself watching an introductory video of about an hour, which she rated as time well spent, and anyone used to customising every step of their process will find it restrictive. On the editorial feedback block her judgement is measured and honest: not deep critique or emotional insight, though it did help her clean up sections faster and spot patterns she had missed.

The complaints land in the same two places every time. First, consistency, which is precisely what the saliency engine claims to solve: a Capterra review states that it loses track of characters, storyline and past events in long narratives. Second, and harsher, a store review describes the program changing major plot points, rewriting important scenes and altering character motivations, and charging credits every time it ignored the instructions given. That last part is backed by the vendor itself, whose 30 June 2026 changelog entry states it fixed failed or partially completed scene output still charging the estimated credits. The same store also carries reports that the subscription management button errors out and blocks cancelling.

That credit detail leads straight to the structural catch, which gets its own section below. Two more things deserve saying first.

The older one still weighs. In May 2023 the tool went through a public storm after observers noted it handled terminology very specific to the fanfiction community, and two outlets that covered it then, The Mary Sue and Distractify, also document a January 2023 programme that asked unpublished authors for complete manuscripts in exchange for free outlines. The company’s answer belongs here too, because it was given and it was specific: it denied scraping anything and attributed that knowledge to the third-party base models it works with. None of that describes today’s product, but it remains part of what you are buying into.

The newer one is worse, because it contradicts in writing a sentence its own documentation still publishes. The page listing the plans states that all of them have full access to all features and that only the credits differ. That has been untrue since May 2026: the changelog declared then that the editorial feedback block became exclusive to the Professional and Max plans, and the feature’s own documentation repeats it. Anyone buying the entry plan believing it carries everything will find the editor missing.

Pricing and plans

The entry plan is called Hobby & Student and costs $19 a month. That is the real monthly figure and it needs stressing, because the pricing page does not lead with it: the toggle opens on annual and what you see on arrival is $10 a month, which is what twelve months paid up front works out to. Two separate checks agree on this, and the company’s own documentation spells it out when listing the three tiers. The ladder continues with Professional at $29 a month and Max at $59, shown as $22 and $44 in the annual view.

Every plan is the same tool with a different bag of credits: 225,000 on entry, one million in the middle, two million at the top. And here is the genuine small print. The company admits there is no fixed relationship between words and credits, because spend depends on how much text is read, how much is written and which model does the work. Its single official example is damning for how wide it is: that one million credits on the middle plan might yield anywhere between fifty thousand and five hundred thousand generated words a month. A tenfold gap between best and worst case means nobody can budget a novel in advance, and it explains without further analysis the Capterra complaint about burning the allowance in a week.

On leftovers, monthly credits only roll over on the Max plan, for up to twelve months, and credits bought separately never expire on any plan. With small print worth reading whole: spending them requires an active subscription, so never expiring does not mean still usable once you cancel.

On the free trial the company contradicts itself. The pricing page carries a button saying start the free trial, with the note that no card is needed and you can cancel anytime, and the home page repeats that it is $10 a month after the free trial. Its help article titled precisely whether there is a trial, updated February 2026, describes none: it describes a refund, saying that if you sign up and decide it is not for you, tell them within the first seven days and they will refund you. With two versions on the table, the prudent one is the second.

And there is a worse bill hiding. Subscribing through the iPhone app rather than the web costs a great deal more: $24.99 for the entry plan, $39.99 for the middle one and $84.99 at the top, against $19, $29 and $59. Same tool with the store’s cut on top, and the gap widens as you climb.

Who it is for (and who it is not for)

It fits anyone writing a novel or a series who wants the AI inside the process rather than in a separate tab. It fits particularly well for writers with a large world who would rather not administer it by hand, since the saliency engine and the series folder are exactly that, automatic context administration. And it fits anyone writing dark genre material, because its own model will not refuse it.

It does not fit anyone who wants to control precisely what goes to the model on each call, which is what Novelcrafter offers in exchange for bringing your own key and building the codex yourself. Nor anyone who needs to know the monthly bill in advance, where the credit bag works against it and DreamGen is more predictable. And it does not fit anyone whose main need is order, genre templates and a free tier to poke around in without paying, ground where Squibler competes better even though its prose is rated worse. The segment ranking sits at creative writing and the wider map at best AI writing tools.

Alternatives to Sudowrite

The comparison turns on who assembles the context. Sudowrite does it for you and does not show the seam, and in exchange you start fast and with good prose. Novelcrafter hands you the whole seam and charges for it in configuration hours plus a separate API bill. DreamGen sits at the cheap midpoint, model included and less apparatus around it. If you recognise yourself wanting to write this afternoon rather than build a system, Sudowrite puts the least friction in the way.

Frequently Asked Questions

Why does the page say $10 while you publish $19?

Because the pricing page opens with the toggle set to annual. The $10 is a twelve-month subscription spread across months and the $19 is what you pay billing month to month. The company’s own documentation lists all three plans with both figures, so there is no ambiguity, only a presentation order that favours the annual deal.

How many words does a plan actually give me?

There is no way to know in advance. The company says spend depends on text read, text written and the model in use, and its only official example puts the one-million-credit plan somewhere between fifty thousand and five hundred thousand words a month. If you generate a lot and discard a lot, plan for the low end of that range.

Does it really not forget the characters?

It has three documented mechanisms to prevent it, and on paper they are the best in the segment: selective retrieval from the story bible, chaining up to twenty-five chapters back, and characters shared across books in a series. That said, user reviews claim the opposite on long works, so the promise holds up better the tidier your Story Bible is.

Will it write explicit or violent scenes?

Yes. Its in-house model is declared as carrying virtually no filter and the documentation says it spans everything from dark and gritty to light and funny, graphic themes included where other models refuse. That is a product decision rather than an oversight, and worth weighing if children use the same machine or if you would rather have guardrails.

Can I use models other than theirs?

Yes, and the catalogue is long. Alongside its own model it offers families from OpenAI, Anthropic, Google, DeepSeek, xAI, Moonshot and Mistral, marked experimental, and its changelog shows recent movement in that list. You do not bring your own key, so everything is paid from the same credit bag.