Disclosure: Type.ai reached out and gave me a free account with unlimited credits so I could review it. I wasn’t paid, there are no affiliate links in this post, and Type.ai had no editorial control. They didn’t see this post before it went live. I also opened a separate free account under another email to see what an ordinary user gets.
I used Claude, Anthropic’s AI assistant, to help plan the tests, keep my run log and draft this post from my notes. The opinions are mine.
The short version
Type.ai is a friendly, easy-to-learn AI writing workspace. It feels like a cross between Microsoft Word or Google Docs and ProWritingAid, with a chat assistant sitting beside your manuscript.
I’m not its target customer. I like local, self-hosted tools and hands-off generation. Type.ai is built for writers who want to stay involved in every step, and for them I think it’s a good fit.
Its biggest problem is objective and fixable: light grey text that falls below the common accessibility standard, and no way to change the font size. Type.ai’s own FAQ says those options are coming.
Its continuity check found real errors in my AI-written novel but missed others, so “the details never slip” oversells it. Every AI check I tried missed something, which is why a human still needs to read the book.
What I tested
I ran three tests between October 10 and 12, 2026, on Type.ai version 2.2.10, then took a second look at the Reviews and audio tools on October 13.
- A 30-chapter novel from my control prompt. This is the Affirmation Glitch prompt I use as a control in my own AI writing benchmarks. I set the preferred model to Claude and the chat to Agent mode. Type.ai offered to outline first, then drafted chapter by chapter while I accepted its suggestions.
- A human-written control. Pearl S. Buck’s 1930 novel East Wind: West Wind entered the US public domain on January 1, 2026. I uploaded a scanned copy, had Type.ai clean it up, then asked it for a continuity check.
- A short story on a free account. I used Type.ai’s guided starter to write a children’s story about my three cockatoos exploring St. Louis. Then I checked how many credits it used.
I chose East Wind: West Wind fairly randomly. Pearl S. Buck is my favorite author after Kurt Vonnegut. I’m interested in early 20th-century Chinese history, and I love that she writes about women and ordinary people. She was also a kind of disability advocate, because her daughter had special needs. Here is the scanned copy I used, on the Internet Archive.
To judge the results, I used ChatGPT with my 10-dimension manuscript rubric (each scored 1 to 5, 50 total; the instructions are in the appendix), AutoCrit’s score and beta reader, and my bookdiff script for repetition. And my own eyes.
Read the unedited output yourself: Type.ai Affirmation Glitch novel page on hellojessicamurray.com and cockatoo story page on hellojessicamurray.com. Both still have the errors I describe below.
The 30-chapter novel
Type.ai finished a 30-chapter draft in 72 minutes, counting my water and bathroom breaks. Microsoft Word counts it at 116,468 words: 129% of my 90,000-word target, with chapters from 3,017 to 5,301 words.
| Test | Who wrote it | Words | ChatGPT rubric (out of 50) |
|---|---|---|---|
| East Wind: West Wind | Pearl S. Buck, 1930 | about 70,000 | 43 |
| The Affirmation Glitch novel | Type.ai, Claude setting | 116,468 | 30 |
| Cockatoo short story | Type.ai, free account | 2,705 | 26 |
ChatGPT called the novel a “rough draft with a viable novel inside it” and estimated 100 to 140 hours of developmental editing. It scored concrete detail highest and freshness of language and continuity lowest. AutoCrit gave it 80.6, though AutoCrit mostly measures sentence and paragraph shape rather than story. Its beta reader gave 4 stars, praised the opening and found the middle repetitive.
Repetition was low: bookdiff found only 0.3% of 10-word passages repeated.
It is not automatic. Each chapter took 3 to 4 minutes, then waited for me to approve it and ask for the next. I had to babysit it. That’s a plus for writers who want control, and a chore across 30 chapters.
Continuity: helpful, but not “never slip”
Type.ai’s ads say it keeps track of your whole story so “the details never slip.” On my novel, its continuity check flagged three problems and missed seven that ChatGPT or AutoCrit’s beta reader found.
After chapter 30, Type.ai offered a continuity edit, and I ran it. Later I asked for “one final continuity pass.” Here is what its check flagged, next to what ChatGPT and AutoCrit’s beta reader flagged.
| Problem in the novel | Type.ai’s check | ChatGPT | AutoCrit beta reader |
|---|---|---|---|
| Maya’s apartment is on the fifth floor, then the third, then the second | Not flagged | Flagged | Flagged |
| Maya’s years in her career are given as 6, 8, 10 and 12 | Not flagged | Not flagged | Flagged |
| Her server has one CPU core in chapter 3 and four in chapter 4 | Not flagged | Flagged | Not flagged |
| She hand-codes the site in chapter 1, but chapter 29 blames a WordPress template | Not flagged | Flagged | Not flagged |
| The site’s address changes between .space, .com and .org | Not flagged | Not flagged | Flagged |
| Characters mention the book’s own chapter numbers (“In Chapter 21…”) | Not flagged | Flagged | Not flagged |
| A five-word prompt is described as “six words” | Not flagged | Flagged | Not flagged |
| Julian, a cellist, carries a violin from chapter 23 on | Flagged | Not flagged | Not flagged |
| A Thursday storm follows a Saturday scene | Flagged | Not flagged | Not flagged |
| Dates and timestamps drift in the later chapters | Flagged | Flagged | Flagged |
I checked the apartment floors by hand. All three numbers are in the book. Separately, my bookdiff script counted 10 places where characters or narration mention the book’s own chapter numbers.
The short story told the same tale on a smaller scale. Type.ai’s check and ChatGPT each found three problems, and they agreed on only one: the prize changes from one golden ribbon to three blue-and-gold ribbons. Only Type.ai noticed my cockatoo Sara sitting on my arm in one paragraph and under the café table in the next. Only ChatGPT noticed the bird carrier with padded straps turning into a stroller.
On Pearl S. Buck’s novel, both found no real continuity problems, and ChatGPT scored its continuity 5 out of 5. Type.ai correctly flagged scan damage, like broken chapter headings and “fect” for “feet,” plus two small timing ambiguities. It didn’t invent problems.
My takeaway: when a book is clean, AI checks agree. When it has errors, each check finds a different handful. In my experience across several tools and models, that’s a limit of how today’s AI handles long text, not something unique to Type.ai. Type.ai’s check is worth running, and it caught errors the others missed. But run more than one check, and read the book yourself.
What I liked
The interface is the best part. Even late at night and tired, I could find everything, and anyone who has used Word or Google Docs would probably feel comfortable sitting down and using all the features. It is super intuitive!
- The rewrite menu. Highlight text and choose Improve, Fix Grammar, Shorten, Lengthen, Simplify or your own instruction. It works a lot like ProWritingAid. I barely used it and want to spend more time there.
- You approve every change. Edits wait in the document as pending, with Accept All, Reject All and controls for each change.
- A guided start. It asks a few questions about your story, then offers to outline first or start drafting.
- The Notes panel. It tracks characters, places and key moments, and it updated as new people appeared in the prose. My novel’s character list grew from 6 to 12.
- Import and export. It imports documents up to 150,000 words and exports PDF, DOCX, HTML and Markdown. When I asked it to format for Kindle, it added a title page, evened out the headings and told me to finish in Word or Kindle Create.
- Privacy. Type.ai’s FAQ says neither Type.ai nor its AI providers train models on your writing.
A second look: Reviews and audio
On October 13 I went back and nearly missed what may be Type.ai’s most powerful feature. The middle icon at the top of the right-hand panel opens Reviews. The icon is faint, and I’d assumed it was basic spell check.
Reviews runs a pass over your draft. You can choose Custom (your own instructions), Improve (remove clichés, fix errors, improve transitions), Fix errors, Shorten, or Enhance structure and formatting. It returns a list of suggestions, each shown as a tracked change with Accept and Dismiss buttons. It’s similar in spirit to AutoCrit or ProWritingAid, but it works line by line with AI rather than producing a report.
For writers who want control over every change, this is great. On a 116,000-word novel, the list gets long. Some suggestions were small fixes, like “ribcage” to “rib cage.” Others were style choices, like cutting “the better part of” from “the better part of six years.” One suggestion struck out four paragraphs at once, which is exactly why you’d want to read each one.
A few rough edges:
- The difference between Improve and Fix errors wasn’t clear to me in practice.
- A pass over the whole novel takes a long time, and there’s no Stop button once it starts.
- The suggestions can’t be exported as a report to read as a whole.
- While Fix errors ran, my Chrome tab used about 1.8 GB of memory and froze. I had about 100 tabs open, so my browser may share the blame.
The audio tab narrates text you select. I counted about 20 voices, from narrative storytellers to ASMR voices, a cowboy and Santa Claus, plus a quality setting. You can download the result: a 268-word passage became a 2-minute-10-second MP3. Audio uses credits.
I could happily spend a few more hours here. With a Stop button, an export, and better contrast and text size, Reviews could be the feature that sells Type.ai.
What I’d change
Contrast and text size come first. The grey labels in the chat panel measured about 3.4 to 1 against the background, while the WCAG AA standard asks for 4.5 to 1 for normal text. There’s one font and no resizing. By late evening my eyes were tired, and the small grey text kept getting harder to read. Type.ai’s FAQ says these options are coming, and I hope that’s soon: it is probably not a big job, and it matters to a lot of writers.
The rest are smaller:
- Let me export the Notes panel. The characters, places and moments can’t be exported, which is a real loss for planners.
- Summarize what the continuity edit changed. The chat lists 20 or more “Searched, Read, Edited document” steps with no description, and many changes were a single word.
- Add an auto-continue option for long drafts, so I don’t have to approve and prompt every chapter.
- Add a “jump to latest” button in the chat panel. Once I scrolled up to review, it was hard to find my way back.
- Name the model. Settings offer Claude or ChatGPT but not which version. Older release notes named models; recent ones don’t.
- State a retention period. The privacy policy doesn’t say how long your documents are kept or how to have them deleted. You can ask at privacy@type.ai.
Credits and pricing
My short story used 8,858 credits on the free account, which suggests a full novel needs at least 380,000. Type.ai’s plans, as listed on its pricing page in October 2026:
| Plan | Price per month, billed monthly | Credits per month | Pitched for |
|---|---|---|---|
| Free | $0 | A fixed amount that doesn’t reset | Trying it out |
| Basic | $12 | 500,000 | Shorter pieces and stories |
| Pro | $24 | 1,500,000 | Novels and screenplays |
| Max | $96 | 6,000,000 | Multiple books a year |
Annual billing takes 33% off. Credits pay for AI features like chat, images and audio. Writing by hand, importing, exporting and sharing are free.
My free account started with 50,000 credits, labelled “Monthly” in the meter, although the pricing page says free credits don’t reset. The guided questions, outline and 2,705-word story used 8,858 of them, about 3,300 credits per 1,000 words.
At that rate, a 116,000-word novel would cost about 380,000 credits. That’s a floor: in a long book, each new chapter has more story for the AI to read, so the real cost is likely higher. On that rough math, Basic covers at most one long draft a month and Pro up to about three, depending on how many revisions and what changes you make. Max’s “multiple books a year” looks realistic. One short story is a thin basis, so treat these as estimates.
Who Type.ai is for, and who it isn’t
Type.ai suits writers who want AI to help rather than take over.
It’s a good fit if you:
- want to approve every change the AI makes
- prefer talking through your story to fiddling with settings
- are new to AI, or a little skeptical of it, but comfortable in Word or Google Docs
- currently paste chapters into ChatGPT and want something built for books
I also think Type.ai may be overlooking older writers who want to try AI while keeping control. The familiar interface suits them. Better contrast and larger text would make it far easier on their eyes.
It’s probably not for you if you:
- want hands-off, automated book generation
- prefer local, self-hosted or bring-your-own-key tools, as I do
- need high contrast or large text today
- want formal analysis reports with statistics on things like passive voice and pacing, the kind AutoCrit provides. Type.ai’s Reviews give line-by-line suggestions instead
Limitations of this test
This was one person testing one version over three days, so read the numbers as one data point.
- I ran the novel on the Claude setting only and skipped the ChatGPT setting.
- AI output changes from run to run. Another run of the same prompt would come out different.
- I forgot to use my control prompt for the short story. It shows how a typical user might start, not a controlled test.
- I didn’t test Type.ai’s style feature or writing in my own voice, because I forgot to fill in the style field. Even so, the novel followed the tone and voice instructions in my prompt from start to finish, which impressed me.
- The judges are AI too. ChatGPT may recognize Pearl S. Buck’s novel, and it scored the raw scan rather than Type.ai’s cleaned-up copy.
- The credit estimate rests on one short story on a free account.
- The novel’s word count is 121,728 with the outline Type.ai left at the top, and 116,468 after I asked it to remove the outline.
- This was a stress test. If I’d been an ordinary user, I would have stopped running The Affirmation Glitch around chapter 3. Most people won’t push it this hard.
Bottom line
Type.ai is polished, friendly and upfront about being a hands-on tool. It isn’t for me, but I can picture plenty of writers who’d be happy working in it, especially those who want AI as a helper rather than an author.
Fix the contrast and text size, give Reviews a Stop button and an export, and it will work for many more of them. And whatever tool you use, read your own book before anyone else does.
Sources
- Type.ai: pricing page, FAQ, privacy policy and release notes, read October 10 to 12, 2026
- WCAG 2.1, Understanding contrast (minimum)
- East Wind: West Wind (1930) on the Internet Archive
- bookdiff and raw data on GitHub
Appendix: my ChatGPT judging instructions
These are the instructions I gave ChatGPT as a judge. I removed front and back matter before running it.
It acts as a developmental editor scoring an unpublished manuscript for a benchmark study. It must not speculate about who or what wrote it. It assumes front and back matter have already been removed. Its response must contain exactly four parts, in this order:
- Coverage check
- Report the number of chapters and the first and last chapter numbers.
- Give the first six words of chapter 1, the midpoint chapter, and the final chapter.
- Identify the longest chapter and its approximate word count.
- If it cannot see the file’s ending, disclose that and stop without scoring.
- Scores. Score these ten dimensions from 1 to 5, with a one-sentence justification in a pipe-delimited table: scene vs summary; forward movement; character want and obstacle; concrete specificity; dialogue and other characters; earned emotional payoff; freshness of language; continuity; onboarding and clarity; would keep reading. The scale defines 1 as unpublishable, 3 as a rough draft with a real book inside it, and 5 as ready for a copyedit. It must use the full range.
- Evidence
- Identify three chapters supporting the lowest score, with one quotation of no more than fifteen words from each.
- Quote any phrase or image recurring at least three times and estimate its frequency.
- Identify the strongest chapter and explain what it does differently.
- Editing estimate
- Estimate the developmental editing hours needed to reach a publishable draft.
- Name the single change that would recover the most value per hour.
Additional rules: judge the actual manuscript rather than its concept; do not soften scores for length or effort; do not congratulate; summarize plot only when necessary to justify a score.
Screenshots of Type’s features:
