Editorial policy
How this content is made
A directory is only worth reading if you can trust how it was put together. So here is the standard we hold every page to: what we source, where we use AI, where a human signs off, and how we fix things when we get them wrong.
Evidence over opinion
Every factual claim about a platform, its price, its voice count, a compliance certification, links to a dated source. Pricing is captured from the vendor's own pricing page on the day it is edited, with the capture date shown. If we cannot source a claim, we do not make it. Where we state a capability a vendor only asserts in marketing, we attribute it to them rather than presenting it as our finding.
Where we use AI, and where we do not
We use AI tooling to help draft and structure pages, the same way a writer uses a research assistant. It does not get the final word on anything that matters.
- AI never invents a number. Prices, latencies, voice counts and scores come from a source or a measurement, never from a model's guess. Performance numbers are produced only by the test rig.
- AI does not decide rankings. The order on a page follows the scoring rubric and the underlying data, not a model's preference.
- No AI-humanising tools. The writing voice is built from real, verifiable samples, not run through a tool that disguises machine text. We would rather sound like a person because the words were written like one.
- A human reviews before anything publishes. Drafts are proposed for review; a person reads, checks the sources and approves before a page goes live. Nothing reaches the site unreviewed.
The performance scores, honestly
The 1–10 scores on the site today are an editorial preview: our provisional, clearly-labelled opinion to get the framework in place. They are not yet from blind test calls, and any latency figure is a mock placeholder. When the test rig has run, measured results replace the preview and the methodology page carries the dated detail. We label the gap rather than pretend it is not there.
Independence
Rankings are never for sale. Some links are affiliate links that earn a commission at no cost to you, and every programme is disclosed in full on the affiliate-disclosure page; affiliate income never moves a ranking. If we ever sell placement, it will be a separate, clearly-marked slot that sits outside the rankings. We hold no equity in and take no fee from any platform reviewed here beyond those disclosed affiliate links.
Corrections
If we get something wrong, we fix it on the page and note what changed, rather than quietly editing it away. A correction that affects a published figure updates the page's modified date and carries a short note. If you spot an error, email hello@voxrater.com and we will check it against the source.
Freshness
Stale data is wrong data, so each page is kept current on a clear cadence:
| When | What happens |
|---|---|
| Every 90 days | A scheduled review of each platform's pricing, voice counts and compliance, so figures do not drift out of date. |
| Over 120 days since the last test | The page either gets a fresh test or carries a visible 'last tested over 4 months ago' notice. We never hide the age of a result. |
| On a material change | A new model, a pricing change or a latency claim triggers a sooner re-check of that platform, ahead of the 90-day cycle. |
| On a dispute | If a vendor challenges a result, we re-check within 14 days and publish both the old and the new figure with the date it changed. |
Who stands behind it
Every review is sourced, dated and signed off before it publishes. Voxrater is an independent benchmark: we place the test calls, score against a published rubric, and stand behind every number. If something is wrong, we correct it in the open, with the date it changed.