A term starts rising. You publish the page and the ad that answer it.
Trend tools tell you what is rising, not what to write. Generic writers have no demand signal behind them and no idea what your product can and cannot do. Trendwright aims at the exact query people are typing — as an article, as Facebook and Instagram copy, or both off the same term — and checks in code that it actually uses their words.
Powered by internet search and AI assistant demand — Google and Perplexity.
Rising demand in, a page aimed straight at it out
What is rising now
Search demand and AI-assistant demand, filtered against an intent lexicon, ranked by momentum. Your backlog sits underneath as a floor, so a dead feed still answers — but the rising terms lead.
Your site profile
What you sell, who you serve, what you may assert about yourself and what you may never imply. The claims policy is injected into the writer, injected independently into the reviewer, and checked in code.
On the term, or nothing
The keywords are checked in code before a reviewer is paid to look. Then the reviewer scores whether it answers that exact search. The publish decision is arithmetic, not a model's opinion — and a draft that misses costs you nothing.
The only idea in the system
Aim at the trend. Use the words people are typing.
A rising term is a population searching one specific phrasing right now. The article’s job is to be the thing that answers that phrasing before anyone else does — so the term goes in the title, in the opening sentence, and in the body. Answer the tidier, more general version of the query instead and you have written a page for a search nobody is running.
| What is rising | Adjacent article | On the term |
|---|---|---|
| “fafsa verification form” spiking | How to Fill In Forms Online | How to Sign the FAFSA Verification Form Without Printing It |
| “hurricane claim photos” rising | A Guide to Insurance Paperwork | How to Attach Photos to a Hurricane Claim Form |
| “acme new pricing” breaking out | Choosing Software for Your Team | What Acme’s New Pricing Costs a Three-Person Team |
| “framework v4 build error” climbing | Understanding Framework Versions | Framework v4 Broke Your Build: What Actually Changed |
Every suggestion carries a capture window — days, weeks or months — and every published draft comes back with a review-by date. These pages have a shelf life on purpose. One that owns a term for three weeks and is then retired did its job; hedging an answer into vagueness so it lasts longer is how you lose the click.
One rising term. Two things worth publishing.
The same signal that tells you which article to write tells you what an ad should say this week. They are separate calls with separate checks, because an article and an ad fail in completely different ways.
Articles
A complete draft aimed at the query, in your voice. The keywords are enforced in code, the claims policy is checked three ways, and a draft that answers a tidier question than the one being searched is discarded.
3, 5 or 15 credits by tier
Facebook and Instagram copy
A set of variants that each test a different angle, with headline, primary text, button and a visual direction. Character limits, your permitted claims and the platform families that get ad accounts restricted are all checked before a reviewer is paid to look.
6, 10 or 25 credits by tier
Why ads cost about twice an article
A set is several pieces of copy rather than one, every variant runs the platform-policy checks, and the reviewer is answering a question that a restricted ad account makes expensive to get wrong. The back-end cost is genuinely higher, so the price is. One credit balance covers both — an agent reasoning about two currencies gets it wrong.
A reviewer that was never told what the writer was trying to do
The reviewer receives the finished text, your claims policy, your style contract, your existing posts, and the query it was written to capture. It does not receive the brief’s argument, the signal, the writer’s reasoning, or any per-call instruction a caller passed in. That is enforced in the function signature, not in a prompt — an anchored reviewer rubber-stamps.
It scores and quotes. It never issues a verdict, because a model asked for a verdict drifts toward approval over time. The publish decision is computed in code from the scores below.
Before any of that, the keywords are checked in code: at least one in the title, all of them in the body. A draft that paraphrases the search never reaches a reviewer, because it was never going to be found.
On failure the engine discards and re-authors from scratch, once. It does not revise. Revision loops are how a pipeline talks itself into publishing what the reviewer already rejected.
Publish thresholds
Every dimension must clear its floor. There is no averaging.
- Task utility
- 4 / 5
- Relevance
- 4 / 5
- Factual safety
- 5 / 5Absolute
- Trend alignment
- 4 / 5
- Originality
- 3 / 5
- Style compliance
- 3 / 5
Trend alignment asks one thing: does this answer the exact search, in the opening, in the reader’s words? Factual safety is required at 5 rather than 4 on purpose — a 4 means the reviewer saw something, and anything touching law, tax, health or money is where an unattended pipeline does real damage.
Metered in credits, priced by tier
Suggestions cost 1 credit. An article costs 3, 5 or 15 depending on how much model you asked for; an ad set costs 6, 10 or 25. Anything that fails its checks costs nothing.
Standard
Most sites$79/mo
- 500 credits included
- ≈ 100 articles, or 50 ad sets
- 3 site profiles
No free tier and no trial credits. Credits are consumed in order: monthly allotment, then overage, then prepaid.
What we read, and what we will not claim
Two source classes, both real: search demand from Google Trends, and AI-assistant demand sampled from the Perplexity API at the moment trends are checked. They diverge, and the divergence is the signal — a topic rising in search but thinly cited by assistants is exactly the whitespace a citable article can take.
Results may be up to a day old, so nothing here is described as real-time or live — which is also why a suggestion you sat on for a week is aimed at a rise that may be over. Query construction, the intent lexicon, sampling cadence and scoring weights stay private: naming providers is a credibility claim; publishing the join would be handing over the work.
Start where the output quality is actually decided: the claims policy.
Add a site, walk the forbidden list, run a test draft against a term that is rising today, read the reviewer’s quotes, tune, run another. An operator who never runs a test draft will be disappointed by their first agent-generated batch.