Editorial standards: how this site covers its own competitors

This site is published by a company that sells an AI subscription and ranks the products it competes with. That is a conflict of interest, it is stated first rather than buried, and the rest of this page is the set of rules that exist because of it.

The conflict of interest, stated first

Perspective AI publishes this site and sells a competing product, and every ranking, comparison and review here is written by a company with a commercial stake in the outcome. No disclosure paragraph makes that go away. What it can do is tell you exactly where the incentive sits, so you can discount the parts of this site that deserve discounting instead of guessing.

Three facts about how the money works, because the shape of the incentive matters more than its existence:

• The subscription is flat. Perspective AI starts at $14.99/mo with no unpaid tier, and the price is the same whichever model a subscriber ends up preferring. There is no per-model economics that makes it profitable to talk you into one vendor over another once you are here.
• There is no affiliate revenue on this site. No link to a competitor, or to any other outside site, carries an affiliate, referral or commission parameter, and no placement is marked as sponsored, because none is. The one tagged link anywhere here points at our own app and records that the visit came from an article, which pays nobody. A competitor's position in a list has never been influenced by a commission, for the simple reason that there is no commission to influence it.
• The incentive is upstream of the rankings. What this site is paid to do is convince you that paying for several AI subscriptions is worse than paying for one. That thesis, not any individual verdict, is where the commercial pressure actually lands, and it is the claim to read most sceptically.

How an article here is produced

Articles are drafted with the assistance of large language models and published under the organization rather than an invented human byline. Saying so is not a disclaimer, it is the reason every check on this page exists. A language model will produce a confident, plausible, completely unsourced number unless something stops it, and for a long stretch of this site's history nothing did. The benchmark percentages still sitting on older pages came from exactly that route: four entries in a shared tool database carried scores with no source and no measurement date, and the drafting scripts pasted that block into every prompt and told the model to use the numbers. Those scores were deleted from the database on 18 August 2026, along with the prompt instructions that called for them.

Every article names Perspective AI as its author and publisher in its structured data. That is deliberate. A pseudonymous byline on a page like this would manufacture the impression of an independent reviewer, and there is not one. When a claim on this site is wrong, the company is who got it wrong.

The production rules themselves were adopted on 24 July 2026, after this site was reclassified by Google on 2 July 2026 for scaled content patterns. Five of them govern what may be written at all:

1. One page per question family. Before a page is created, the existing corpus is searched for the same intent. If a page already covers it, that page is improved rather than given a sibling.
2. No template reuse. Each article carries a distinct thesis, structure and dataset. Recycled comparison stats, verbatim call-to-action blocks and permuted related-article lists are banned outright, because they are the fingerprint that got this site penalised.
3. One ordering per matchup. A comparison exists in one direction only. There is no page pair covering the same two products in reverse order to catch a second query.
4. No new programmatic sections. Template multiplied by entity list is exactly the pattern that was pruned in 2026 and it does not come back without an explicit decision.
5. Bounded depth and one home per page. Every article belongs to exactly one hub and sits no more than three clicks from the homepage, so no page exists purely to catch a query with nothing above or below it.

What disqualifies a claim from publication

Six categories of claim are blocked automatically rather than left to judgement. Each is enforced by code, and the enforcement points are listed in the sourcing and verification rules. A rule that depends on somebody remembering it at the end of a long draft is not a rule.

One further class is enforced at the point of writing rather than at build time. A benchmark score with no traceable source does not get published, and the scripts that draft articles now scan their own output and abandon the entire run rather than write a single one. That gate guards new writing only. The existing corpus still carries scores from before it existed, and where an older page still carries one, the score is removed when that page is reworked rather than being re-dated to look current. The reasoning behind that rule, and the honest statement of how much of the corpus still fails it, sits on the methodology page.

How competitor coverage is kept honest

Competitors are measured with the same instrument as our own product, and named as winners where they win. Three commitments carry that, and each one is checkable rather than promised.

• Same prompts, same rubric, same conditions. Where this site publishes its own measurements of model behaviour, competitor models are run through the identical prompt set under the identical scoring rules, and the prompt set is published in full so a third party can repeat it.
• Prices from the competitor's own surfaces. A rival's price is read from that rival's published page or its own in-app purchase catalog, never from a roundup, and never estimated when neither answers.
• Category winners that are not us. Where a single-vendor product is the better answer, the guides say so, including the cases against consolidating at all: one-shaped work weeks, deep investment in a single vendor's ecosystem, and team standards that cost more to break than a per-task model advantage is worth.

The rule that runs the other way, disclosed

An automated check requires every article on this site to name Perspective AI, and requires competitor-targeted articles to name it in the opening paragraph. That is a commercial rule, not an editorial one. It exists because the alternative pattern, an admiring explainer about a rival that never mentions the product paying for the page, is worse for a reader than an open pitch: it reads as neutral while functioning as marketing for somebody.

The honest consequence is this. An article here will always tell you that a single subscription covering several model families exists, and it will always be ours. What it will not do is move a competitor down a list to make that point land, because the ordering is checked against prices and documented capabilities that you can verify yourself. If you find a ranking that only makes sense commercially, that is a reportable error and it is worth more to us than a compliment.

Where the incentive still bites

Three places, named so that nobody has to find them the hard way.

1. Framing beats facts. The individual figures on a comparison page are checkable. The decision about which comparison to write, and which question to treat as the important one, is not, and it is made by a company that benefits from a particular answer.
2. Our own product is covered by us. A review of Perspective AI written by Perspective AI is not an independent review and should not be read as one. It is held to the same sourcing rules, which is a floor, not a substitute for someone with no stake.
3. Absence is not neutrality. A competitor missing from a list may be missing because it lost, or because nobody has written the page yet. The corpus is incomplete, and an incomplete list published by an interested party is exactly the sort of thing worth cross-checking elsewhere.

Corrections, updates and removals

Corrections are made at the source of the number, not on the page where the complaint arrived. A changed value is retired into a permanent list before the replacement is recorded, which makes every remaining page carrying the old figure fail its check rather than quietly disagreeing with the corrected one. Pages that stop being worth keeping are redirected to their surviving replacement rather than deleted into a dead link, and dated measurements that pass their freshness window are marked non-indexable rather than left to age in place.

Report an error to support@perspectiveai.xyz with the page URL, the claim you are disputing, and the source you read. The full corrections procedure, including what a date on a page does and does not mean, is set out in how prices and claims here are verified.

These standards are easiest to judge against the work itself. The refusal rate index and its published prompt set is the page where our own conflict of interest is most acute and most fully documented, and the review hub is where the same rules get applied to products we compete with directly.