Methodology of the ranking

The rule every criterion obeys

A criterion scores only when the points can be traced to one sentence published by the vendor, on an address that is printed next to it, on a date that is printed next to it. When no such sentence exists, the criterion scores zero and the tool page prints not documented publicly. Across the 32 tools, 100 of the readings on the seven positive criteria fall that way, which is the honest shape of a market where most vendors publish very little.

Nothing is weighted after the fact. The points below were fixed before the tools were scored, and they are the same for a tool with two thousand customers and a tool with a landing page. Change a sentence on a vendor’s site and the ranking moves on its own.

The seven criteria that add points

• up to +12

  Platforms covered

  The platforms the vendor lists on its own site as places where it talks to fans. A platform its own pages describe as carrying alerts, notifications or a community link is not counted as a chatting platform, and the sentence saying so is printed on the tool page.

  • +12 OnlyFans and Telegram both
  • +6 one of the two
  • 0 neither

• up to +10

  Autonomous selling documented

  Whether the vendor’s own pages state that the AI both sends the message and closes the sale itself, without a person pressing send. Two shapes score 0 and are quoted where they exist: the AI proposes a reply and a person sends it, and the AI opens or warms the conversation but hands the fan to a human for the sale.

  • +10 the AI sends and sells on its own
  • 0 a person sends or closes, or not documented publicly

• up to +8

  Sells inside the conversation, on the platforms it covers

  Whether the vendor documents the paid item being sent and paid for inside the conversation itself, on every platform it sells on. The criterion is platform neutral on purpose: a pay per view sent by the AI in an OnlyFans direct message counts exactly like a locked media paid in Telegram Stars in a Telegram chat, and a Fanvue or MYM equivalent counts the same way. What it separates is not which platform a tool runs on, it is whether the fan pays without leaving the conversation. A sale that leaves the chat, through an external payment link, a redirect to another site or a hand over to a person who closes it elsewhere, scores 0, and so does a sale whose mechanism the vendor does not describe.

  • +8 on every platform it sells on
  • +4 on some of its platforms only
  • 0 the sale leaves the chat, or not documented

• up to +8

  A performance indicator published with its definition

  Whether the vendor publishes a performance figure AND publishes what that figure counts. A number published without its definition scores 0 and is named on the tool page, because a ratio whose base is unknown cannot be compared with anything.

  • +8 a figure and its definition
  • 0 no figure, or a figure with no definition

• up to +6

  Public price

  Whether an amount or a rate for the paid offer is printed on the vendor’s own page. Pricing on request, a 404 at the pricing address, or a commission whose rate is not published all score 0.

  • +6 printed on its own page
  • 0 not published

• up to +6

  Public documentation consultable before buying

  Whether a documentation site, a help center or a public guide can be read without an account and before paying. A FAQ section, a blog, a machine readable file or a sitemap is not documentation, and where that is the case the sentence saying so is printed.

  • +6 published and readable before buying
  • 0 none published

• up to +5

  Hybrid mode or human takeover documented

  Whether the vendor documents that the AI runs the conversation AND that a person can take it back, per chat or per plan. A tool whose AI only proposes replies scores 0 here: there is nothing to take back, since the person never left.

  • +5 documented
  • 0 not documented publicly

The three criteria that take points away

These are the reason a tool with a wide feature set can end up behind a small one. A penalty is never an accusation and never a legal conclusion: it records that two pages published by the same vendor say two different things, that a payment path has one option, or that a sourced reservation was compiled.

• down to -6

  Reliability of published information

  Three faults, each read on the vendor’s own pages, each quoted in full with the address and the date, and each cumulative with the others. A statement about another vendor that the other vendor’s own public page contradicts costs 6, and both pages are quoted with both read dates so the reader can see whether the claim was simply overtaken by a change. A contradiction between two of the vendor’s own pages about its price, its revenue share or its model costs 4. A performance or scale figure published without the definition of what it counts costs 3, because a ratio whose base is unknown cannot be checked by anyone. No intent is attributed to anyone, and nothing here is a legal conclusion.

  • -6 a statement about another vendor verified false
  • -4 a contradiction between two of its own pages
  • -3 a figure published without its definition
  • 0 none of the three found

  We read every comparison page published by the 32 vendors for statements about other vendors. The minus 6 line triggered twice, on two different vendors (ChatSan and Desirely), both for a statement about OnlyChat that OnlyChat’s own pages contradict; no other vendor page was found to contain such a statement, including OnlyChat’s own comparison pages, which were read the same way.

• down to -3

  Payment openness

  Whether cryptocurrency is the only documented way to pay the vendor or to be paid by it. A tool that also takes a card, or that also pays out to a bank account, is not penalised.

  • -3 cryptocurrency only
  • 0 not cryptocurrency only

• down to -3

  Documented reservations

  Minus one point per sourced reservation that NO other criterion already reads, capped at minus three. This criterion is deliberately narrow: a commission on every plan is a pricing fact and belongs to the price criterion, a crypto only payment path belongs to payment openness, a hybrid mode reserved to one plan belongs to takeover, an absent price belongs to price, an absent help center belongs to documentation, and a contradiction or an undefined figure belongs to reliability. Only what no other criterion reads is counted here, and the lines counted are printed on the tool page so the arithmetic can be checked. A tool with no remaining line scores 0, which means nothing was left over, not that nothing exists.

  • -3 three or more reservations no other criterion reads
  • -1 one such reservation
  • 0 none left over

One fact, one criterion

A points model is easy to cheat without meaning to. Charge a tool once for taking a commission on every plan, again because that commission appears in a compiled reservation, and a third time because the commission is mentioned in its terms, and the arithmetic stops describing anything. So every sentence is charged to exactly one criterion.

The rule bites hardest on the reservations criterion, which is deliberately the smallest penalty on the page and is now capped at minus 3. A commission on every plan is a pricing fact and belongs to the price criterion. A cryptocurrency only payment path belongs to payment openness. A hybrid mode reserved to one plan belongs to takeover. An absent price belongs to price, an absent help center to documentation, a contradiction or an undefined figure to reliability. What is left over, and only that, is counted here, and the surviving lines are printed on the tool page so the subtraction can be read line by line.

One pair of criteria needs its own sentence, because the objection is obvious. The indicator criterion never subtracts: a tool that publishes no figure scores zero there, exactly like a tool that publishes an undefined one. The three point reliability penalty is what separates the two, and it is the only place an unverifiable number costs anything. Publishing nothing is neutral; publishing a number nobody can check is not.

How ties are broken

Two tools can reach the same total, and several do. The first tie break is the number of platforms the vendor lists on its own site, the second is alphabetical order on the tool name. The result is a full order from 1 to 32, with no shared places and no editorial nudge. A tool that moves one point moves in the list, which is the point of publishing the arithmetic.

Where the facts come from

Three sources, and no fourth. The row of each tool in the shared dataset, written on 2026-09-14 from that tool’s own public pages. The sourced reservations, each one a sentence read on a vendor page with its own address and date. And the platform documents, quoted where a criterion touches them. No listicle, no directory, no press release, no vendor comparison page is used as a source of fact about a third tool, for the obvious reason that a vendor comparing itself with a competitor is not a source about the competitor.

Prices are read on the vendor’s own pricing page or, where there is none, on the page that prints the amount. A pricing address that answers 404 is recorded as a check, not as an absence of product: the tool simply scores zero on the price criterion and the 404 is printed.

WHAT THIS RANKING DOES NOT MEASURE

Three things decide whether a tool is worth its price, and this ranking measures none of them, because no tool in the list publishes them.

• Writing quality. Whether the AI sounds like a person or like an assistant is the whole product, and it cannot be read off a public page. No score here is a judgement of how a tool writes.
• Conversion. How much a tool sells for a given creator depends on the media library, the scripts, the audience and the price floors, none of which belong to the tool. A vendor figure with no published definition is not evidence, which is why publishing the definition is the criterion rather than publishing the number.
• Support. Response time, onboarding, what happens when an account breaks at two in the morning. Not public anywhere, for any tool here.

A high place means a tool publishes more of what a buyer needs to check before paying. It does not mean it is the right tool for a given creator, and this site never says it is.

What a vendor can do about its place

Write to [email protected] with the address of a public page that prints the fact. That is the only currency accepted here: not a claim in an email, not a screenshot of a dashboard, not an offer. If the page says what the email says, the fact is updated, the read date changes, the points are recomputed and the position moves on its own. If the page does not exist yet, publishing it is the fix.

A correction that is refused gets a reason. A place in this ranking has never been for sale, there is no affiliate link on this site and no advertising slot to buy, and the penalties apply to the tool this ranking puts first exactly as they apply to the others. It takes the full minus 3 on reservations for the limits it publishes about itself, it scores zero on human takeover because it does not document one publicly, and its own comparison pages were read for false statements about named competitors the same way every other vendor’s were. In that last case none was found, and the checked and nothing found line is printed on its page with the address and the date, so the reader can see the test was run rather than skipped.

Sources

1. Shared tools dataset: 90 tools, 32 in the AI chat category, every row written from the tool’s own public pages · read 2026-09-14
2. The compiled reservations, one line per fact, each carrying the vendor page it was read on and the date · read 2026-09-14
3. OnlyFans, Terms of Use, read directly to verify that they contain no rule on AI in messages · https://onlyfans.com/terms · read 2026-09-14
4. Telegram, Bot Payments API for digital goods, on what a buyer pays to acquire Stars · https://core.telegram.org/bots/payments-stars · read 2026-09-14