Should I vibe code
SEO/GEO content platform for research, briefs, optimization, and AI writing
You can build the scoreboard. You cannot build the referee — and the score is the thing people obey.
?
Their verdict, the Starter 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
A model with a search API attached will produce briefs that look exactly like Frase's, in an afternoon, and the headings it pulls out will be the same headings. The difference is downstream: Frase's optimisation score has been tuned against ranking outcomes, and yours is a term-frequency count you invented over lunch. That matters because the number is not a report, it is an instruction — a fabricated target will happily march you, or the freelancer you pay by it, through forty pages built to satisfy your own arithmetic.
What actually breaks
not "if". the specific failures.
- SERP acquisition — scraping Google directly is against their terms and breaks in practice; the version that works is a metered API, which turns your free tool into a monthly bill
- The optimisation score, which is a number you made up and then start obeying
- The GEO score in particular: nobody outside the labs knows what makes an assistant cite a page, so any figure you print is confidence with nothing behind it
- Competitor page extraction, which breaks quietly every time somebody ships a new theme
- Content published at volume against the wrong target — slow to notice, slower to undo
Is that you?
the verdict is a default, not a law
- It is research for you: pull the top ten, extract the headings, drop them in one document
- The score is shown as a heuristic with its formula visible next to it, not as a grade
- You pay for SERP data through an API and treat that bill as the running cost of the tool
- You are scraping Google directly at any volume
- Someone else — an agency, a freelancer, a junior — is being measured against your invented number
- It writes and publishes with no human between the two
If you build it anyway
the checklist, then the prompt that enforces it
- Buy the SERP data. A metered API whose terms you have read is the only version that keeps working, and the bill is the honest signal about whether this tool is worth building.
- Print the formula beside the score. If a reader cannot see that it is term coverage against ten pages, they will treat it as truth.
- Cache by keyword and date, and dedupe. The same query twice in a week is money you did not need to spend.
- Never auto-publish. Draft, review, publish, with a person in the middle looking at a diff.
- Keep extracted competitor text out of the output. A brief is research; a brief that pastes someone's headings into your page is plagiarism with extra steps.
- Store the score's inputs with every brief, so when nothing ranks you can see what the tool believed at the time.
I am building an SEO research and content-brief tool. The risk is not security,
it is that this thing will manufacture a number I then treat as truth. Apply
these constraints and push back if I ask you to break them.
1. Do not scrape Google, Bing or any search engine directly. Use a SERP API
with terms I can read. If I insist on scraping, tell me it breaches their
terms, breaks without warning, and gets the IP blocked — then make me
confirm before writing a line of it.
2. Cache every SERP response by keyword and date, and refuse to re-fetch the
same query inside a window I set. Show me the running API cost in the UI.
3. Whenever you display an optimisation score, display the formula that
produced it in the same component. No score without its working.
4. Do not invent a GEO or "AI citation" score and present it as measured. If I
ask for one, say plainly that the inputs to assistant citation are not
public, and label anything you build as a guess.
5. Extracted competitor content is research material only. Keep it in a
separate field from any draft, and never let it flow into generated output.
6. Never publish. The tool produces drafts; a human moves them. No CMS write
credentials, no scheduled posting, no auto-commit.
7. Store the brief with its sources, its scores and the date, so a page that
underperforms can be traced back to what the tool believed at the time.
8. Write briefs to markdown files on disk, one per keyword, so they outlive
this app.
9. When a scraper for a competitor page fails, fail loudly and skip that page.
Never silently score a brief against three sources while claiming ten.
10. Out of scope: rank tracking, backlink data and anything else that needs a
crawl of the whole web. Those are the parts you cannot build, only rent.
11. Tell me the monthly API spend at which paying Frase is simply cheaper.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
As soon as the tool is deciding what somebody else writes. $49 a month buys SERP data you do not have to meter, a score calibrated against ranking outcomes rather than your own arithmetic, and one bill instead of three — and the API spend alone gets close to that faster than you expect.
$49/mo is cheaper than your weekend.
Briefs are documents, so write them out as markdown in a folder — one file per keyword, with source URLs and the date at the top. Do that and the app becomes disposable: every brief you generated stays useful long after you have deleted the thing that made it, and the only ongoing cost is the SERP subscription you can cancel independently.
Self-hosted metasearch engine with a JSON API. Not a brief tool, but it is the honest answer to "where does the SERP data come from" if you will not pay for an API.
Open-source search and answering over a corpus. Not a Frase clone, but prior art for the retrieval-plus-generation half of the problem.
Questions
Isn't the AI writing the hard part?
It is the commodity part. Generating a well-structured 1,500-word draft from an outline is table stakes now and costs cents. Everything expensive about Frase sits either side of it: the search data going in, and a score coming out that has been checked against whether pages actually ranked.
What is actually wrong with scraping Google myself?
Three things, in ascending order of annoyance: it is against their terms, it breaks constantly and silently, and the fix is a SERP API — which means the free version of your tool was never free, it was just billed later. Once you are paying for data, the comparison against $49 looks different.
So the demo is worth building?
Yes, and that is what DEMO ONLY means here. A weekend gets you a research view that pulls ten pages, extracts their headings and hands you an outline — genuinely useful, genuinely yours. The part to throw away is the score, and the instinct to trust it.
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.
Your rounding bug is not a bug. It is a discrepancy on a filing with your name on it.
Crawling your own site and reading the headers is a script, not a licence.
SERP analysis means scraping search results, which is a fragile foundation to build on.
last reviewed 2026-08-04 · verdict is editorial and unsponsored · shared entry data from canivibecodeit under MIT · not legal advice