Editorial standards.

Sourcing

Every non-obvious factual claim cites a primary source. Where we cannot find a primary source, we either omit the claim or label it as a community report. Marketing-blog citations (the artsmart.ai-style content farms) do not count as sources. Vendor press releases are sources for what the vendor said, not for what is true.

Subscriber and user counts that we cannot verify externally do not appear in our writing. We track public subreddit subscriber counts in the Companion Index as a rough proxy; we do not extrapolate from those to total user numbers.

Voice

Direct, informed, slightly opinionated, never sales-y. We write the way a knowledgeable friend would, not the way a brochure does. We avoid throat-clearing, balanced "it's not just X, it's Y" constructions, and the kinds of phrases that read as filler ("delve," "tapestry," "ever-evolving," "navigate the landscape"). We avoid the em dash because it is the most recognizable AI tell.

We use the active voice, prefer specific over abstract, and lean on concrete examples over generalities. When we have an opinion, we say so and explain it. When we are unsure, we say that too.

Affiliate links and disclosure

The Companion Report is reader-supported. When affiliate links are live on a page, that page carries the FTC affiliate disclosure visible above the fold. Standard wording:

The Companion Report is reader-supported. We may earn an affiliate commission when you click through and sign up for products we recommend, at no additional cost to you. Our reviews are independent and we never accept payment in exchange for review position.

We do not accept sponsored content, paid reviews, or guest posts. Vendors may correct factual errors and suggest topics; we will not change a verdict in exchange for either.

What we will not publish

Three categories of content are excluded from the site, even if a reader, vendor, or contributor specifically asks for them.

Content involving minors in sexual or romantic contexts. Period. We do not review apps that allow this. Where we cover such apps, it is to expose them.

Apps that promote real-person likenesses without consent. If an app's main pitch is letting users build AI versions of celebrities, exes, or other identifiable real people, we cover it only to flag the legal and ethical risks. We do not link to or affiliate with such apps.

Anything framing AI companions as treatment for clinical conditions. We can describe what users say has helped them. We can review wellness apps. We do not write that an app can treat depression, anxiety, addiction, or any other clinical diagnosis. The full mental-health disclaimer appears at the bottom of every mental-health-adjacent article.

Methodology disclosure on reviews

Where we have completed thirty days of hands-on testing on a paid account, the review says so. Where the review is based on synthesis (public information, app-store reviews, community discussion, vendor docs, third-party reporting), it says that too. Pending reviews are flagged as pending. We revise as testing completes.

Dates and updates

Every article is dated. The byline shows when the article was first published and when it was last substantively revised. If a piece changes after publication in a way that affects the conclusion, we add a note explaining what changed and why.

Corrections

If we got something wrong, write us at contact form. Corrections are made quickly. Where the correction is meaningful (changes a claim, a verdict, a recommendation), we add a corrections note at the bottom of the article describing what was wrong, what is now correct, and when it was changed.

Conflicts of interest

The Companion Report is independently owned. We have no financial interest in any of the companies we review. If that ever changes (an acquisition, an investment, a paid role for someone on the masthead), it will be disclosed at the top of every relevant page.