Should I vibe code
Cluster keywords, prepare a brief, and manage an article through a structured content workflow
It resells Google NLP at 60 calls for $19. That line is the company telling you which half you cannot build.
?
Their verdict, the All-In-One Basic price and the build-time estimate come from their entry, MIT-licensed. Checked 2026-08-04.
?
Our verdict, the regret score and everything below it. Editorial and unsponsored — nobody can pay to be moved.
The honest answer
why the verdict is what it is
Read the price list rather than the feature list and this entry writes itself. WriterZen no longer sells a subscription at all — it sells a licence once and then sells credits: keyword credits by the ten thousand, Google NLP credits sixty for nineteen dollars. That split is the company telling you, in its own checkout, which half of the product is software and which half is metered supply. The software half is genuinely a few days for an agent. Clustering a keyword list is embeddings and a distance threshold; a content brief is a template over competitor headings; the editor is an editor. The supply half is Google Ads volume data, SERP results, Google's NLP entity analysis and a plagiarism API, and every one of those is billed per call to you at retail with no volume agreement behind it. You will build the impressive-looking part in a weekend and then discover that running it for one client for one month costs more than the licence you were avoiding. Nothing here is dangerous. It is just an expensive way to arrive at a worse version of a $270 purchase, and the recurring cost never goes away.
What actually breaks
not "if". the specific failures.
- The keyword data, on the day Google Ads returns bucketed ranges instead of volumes and half your clustering input becomes '1K–10K'
- The SERP fetch, once you are scraping rather than buying — CAPTCHAs first, then an IP block, then a proxy bill that quietly exceeds the licence you were saving on
- Clustering thresholds, which look excellent on the demo list of 200 keywords and produce one giant cluster and 400 singletons on a real list of 40,000
- The plagiarism checker, which does not exist as something you write — it is a paid index API, priced per document, forever
- Entity and NLP analysis, metered per call by Google, where a single bulk run over a competitor set is a real invoice rather than a rounding error
- The briefs, which encode whatever you believed about ranking in the month you wrote them and then quietly keep asserting it
- The credit accounting you inevitably build to stop all of the above, which is a billing system you did not want and now maintain
Is that you?
the verdict is a default, not a law
- The clustering runs over keyword exports you already have, from a source you already pay for
- It is one person planning their own content, not a service other people submit jobs to
- You have priced a month of API calls at your real volume and the number did not surprise you
- The brief is a starting document a human edits, not an output anyone treats as research
- The plan involves scraping Google search results at any volume worth having
- Clients are paying for the output, which makes a wrong volume figure a bill they can dispute
- Nobody has decided what happens when the keyword API changes its response shape, because it will
- You need the plagiarism check, which is a purchase with a wrapper, not a build
If you build it anyway
the checklist, then the prompt that enforces it
- Price the data before writing the code. One spreadsheet: keywords per month, SERP calls per keyword, NLP calls per brief, at published API rates. If that number is above the licence, you have your answer without a repo.
- Buy SERP data from an API that sells it. Scraping Google is against its terms, breaks without warning and turns into a proxy subscription that costs more than the tool.
- Cache every external response by query and date, permanently. The same keyword looked up twice in a week should cost once, and a cache is also your only defence when a provider disappears.
- Treat clustering as a tunable, not a truth. Expose the threshold, show cluster sizes, and make it trivial to re-run — a silent parameter that produces one 9,000-keyword cluster looks exactly like a working system.
- Record the source and date on every number that reaches a brief. Volume figures age badly and 'where did 12,000 come from' is a question you will be asked in a meeting.
- Put a hard spend cap on every metered API with a refusal at the limit, not an alert. Bulk operations are exactly where an accidental loop becomes a four-figure invoice.
- Keep the writing tool and the research tool in separate processes. The editor should stay usable on the day the keyword provider is down, which will be a day.
I want to build an SEO content workflow: keyword clustering, a brief
generator, and an editor. Before any of it, I need to be honest about which
parts are code and which parts are purchased data. Order the work that way.
1. Start with a cost model, not a feature. Ask me my expected monthly volume,
then produce a spreadsheet of what keyword volume, SERP and NLP calls cost
at published API rates. Show me the total before writing application code.
2. Do not scrape Google. If I ask, refuse and explain: it violates the terms,
it breaks without notice, and the proxy bill exceeds the tool I am avoiding.
Offer paid SERP APIs with their prices instead.
3. Every external call goes through one client with a cache keyed by query and
date, persisted to disk, never expiring automatically. Build the cache
before the first live call.
4. Hard spend caps per provider, enforced by refusing the call. Not warnings.
Bulk operations are where an accidental retry becomes a four-figure invoice.
5. Clustering comes next and works on a CSV I supply — no network at all. Use
embeddings plus an explicit, configurable distance threshold.
6. Show me cluster count, size distribution and the ten largest clusters after
every run. A silent one-cluster result must be visible, not inferred.
7. Every figure that reaches a brief carries its source and the date it was
fetched. Briefs are documents with provenance, not dashboards.
8. The editor is a separate process from the research side and must remain
fully usable when every API is down. Prove that with the network disabled.
9. Plagiarism checking is out of scope. If I ask, tell me it is a paid index
API priced per document and not something to reimplement.
10. No credit system, no per-user metering, no accounts. If I ask for those,
point out I am rebuilding the vendor's billing rather than their product.
11. Finish by comparing my modelled monthly API spend against a one-time
licence for an existing tool, and say which one you would choose.That one keeps you out of trouble. For the prompt that actually builds it, canivibecodeit.com has one.
their build prompt ↗Or don’t build it
the boring option, and the way back out
Almost always, and unusually cheaply: WriterZen no longer charges rent. All-In-One Basic is $270 once, and even a modest month of keyword and NLP API calls at retail will approach that. Buy it — or any of its neighbours — when you want the data pipeline and the clustering to be somebody else's problem, which is most of the time. Build only the clustering, on data you already pay for, if you want a specific grouping their model refuses to produce. And note what you are buying: a company that has swapped recurring revenue for one-time licences and top-up credits, whose AI assistant is still advertised as GPT-4o-mini, has told you which way it expects the product to develop.
The durable asset is the cached data, not the app: keyword responses, SERP snapshots and NLP results stored as dated JSON, one file per query. Keep those and any future tool can be fed from them, including a spreadsheet. Keep clusters and briefs as Markdown or JSON in the same tree so the research history survives a rewrite. What does not survive is anything you scraped — it has no provenance you would want to explain, and it should not be the thing a client decision rests on.
Actively developed interface for local and API-backed language models, covering the drafting half of this workflow.
Embedding-based keyword extraction and grouping, which is the clustering feature minus the packaging.
Questions
Is WriterZen still being developed?
It is operating and you can buy it today, but the signals point at a product in a holding pattern: the subscription has been replaced by one-time licences, the AI assistant is still described on every page as GPT-4o-mini powered, there is no changelog, and the promotional banner is a New Year campaign still running in August. None of that makes it a bad purchase at $270. It does mean you should not assume the feature list will grow while you own it.
Isn't keyword clustering the hard part?
It is the visible part, which is not the same thing. Embedding a keyword list and grouping by distance is an afternoon and any agent will do it well. The difficulty is upstream: getting forty thousand keywords with volumes attached, and downstream: a threshold that produces useful groups at that scale instead of one enormous cluster. The first is a purchase and the second is tuning you can only do with real data.
What about the plagiarism checker — surely that is buildable?
Not in any meaningful sense. Checking whether a passage appears elsewhere on the web requires an index of the web, which is why every product in this category resells someone else's API by the document. You can build the interface in an hour. The thing behind it is a permanent per-check cost, and it is the clearest example of the pattern this whole entry is about.
Every week, someone ships something they shouldn’t have.
New verdicts, the worst thing that landed in the trap, and the occasional incident report. No other email, ever.
Drafting with sources is a prompt chain. Verifying the sources is the part people skip.
Building the generator takes an afternoon. Building a way to tell when it got worse is the actual product.
Comparing marketing variants is a prompt loop and a scoring rubric you invent.
last reviewed 2026-08-05 · verdict is editorial and unsponsored · shared entry data from canivibecodeit under MIT · not legal advice