SMF·RYTR
Can a prompt replace Rytr?
AI writing — AI drafting, research and content optimization
Exhibit tracking slip
Verdict
Rytr's shape is a variant machine: pick a use case, set a tone, get several versions, keep fragments from different ones. Building the generation loop is trivial; building the interface that makes comparing six variants fast rather than tedious is the part that takes days, and it is the whole product. There is no dataset or infrastructure behind Rytr to be missing — what you are really weighing is nine dollars against your own API bill and your own weekend.
Exhibit A — The prompt
Received on31.07.2026Build a short-form copy generator organised around comparing variants.
Use cases: a use case is a named template holding an input schema (the fields the user fills — product, audience, key benefit, constraint), an output schema (a headline, a paragraph, a list of three), and a prompt built from both. Ship no use cases; the first run creates one from a real task the user describes, and use cases are stored as files so a set can be shared.
Tone, as structure rather than an adjective. Three independent controls, each with a small fixed set of values, composed into the prompt explicitly:
- Formality: casual, neutral, formal
- Energy: understated, plain, enthusiastic
- Person: first person singular, first person plural, second person
A named tone preset saves a combination plus a short style note. This beats a single free-text "tone" field because the axes are independent and a user can change one without disturbing the others.
Variant generation, which is the point of this build. One request returns N variants (default six) in a single structured call — not N sequential calls, because the latency of six round trips is what makes this tedious. Display them in a grid, all visible at once without scrolling where the output is short.
The comparison interface, which is where the effort belongs:
- Select any line or phrase from any variant and send it to a composition pane; the final copy is assembled from fragments across variants, which is how people actually use these tools.
- Pin a variant to keep it visible while regenerating the others.
- Regenerate only the unpinned slots, so a good result is never lost to a reroll.
- A diff view between any two variants, so near-identical outputs are visibly near-identical rather than read twice.
- A discard action that records what was rejected, and an option to feed the rejected set back into the next request as things to avoid.
History: every generation stored with its use case, inputs, tone, model and all variants, searchable, so a good line from three weeks ago can be found again.
Use Anthropic or OpenAI, whichever key is present, selectable per use case. With no key, the history, use cases and composition pane must still work.
Out of scope: a browser extension, long-form article generation, SEO scoring, and any hosted service or account.Opening prefills the prompt — press enter to run it.
Exhibit B — What you lose
- B.1 the breadth of ready-made use cases and their tuned prompts
- B.2 the browser extension that generates wherever you are writing
- B.3 language coverage beyond what your chosen model handles well
- B.4 a fixed monthly price instead of per-token billing
Prior art
Exhibit C — Why people still pay: workflow, data, and model tuning
Because nine dollars is genuinely cheap, and at that price the honest comparison is against your own time rather than your own API spend.
Questions
Can I bring my Rytr use cases across?
There is no export, so each use case is re-created — which is quick, since a use case is a handful of input fields and a prompt. Your saved outputs can be pasted into the history as text if you want them searchable.
Is this actually cheaper than nine dollars a month?
For light use, yes: six short variants is a few hundred tokens out, so a cent or two per generation. Generating fifty times a day gets you to roughly the same place as the subscription. The honest reason to build it is control over the prompts, not the money.
Why generate six variants instead of one good one?
Because short copy is a search, not an answer. The first result is rarely the best and the useful output is usually a line from one variant and a line from another, which is why the composition pane rather than the generator is where this build's effort goes.
What is the one thing that does not survive the rebuild?
Breadth. Rytr ships dozens of use cases already written and tuned, and yours starts empty. The tool only becomes useful after you have written the five templates you actually need, which is an evening most people would rather spend not doing that.
Related tools
Receipt