How the Reply Guy Rankings measure which brands answer on X

Brandon MarshallNovember 20, 2026
“How the Reply Guy Rankings are measured”, under the Vibewatch wordmark, above the Vibewatch mark beside the X logo.

The Reply Guy Rankings rank web3 brands by one number: of the posts on X that needed a reply from them, how many got one in public. One brand's finished score is published each week, measured over that brand's own 30 days. This page is how that number is made, so anyone can check a score and anyone ranked on it can argue with it.

A post that needed a reply is a post that tags the brand's main or support account, or replies directly to one of its own posts, and asks a question, raises a problem, gives feedback or gets something wrong, where our scorer judged an answer would help. Praise counts only when it's personal and a reply clearly matters.

An answer is a public reply from the brand's main or support account or any account on its X affiliates list, within 72 hours of the post. For a reply under one of the brand's own posts, the answer has to be to that reply.

The same rules apply to every brand we rank.

What counts as a brand

A brand is its main X account, any support accounts it runs, and the accounts X lists under its Affiliates tab, such as staff, regional and product accounts. Posts that mention the main account or a support account are what we read, replies included. A reply under an affiliate's post reaches us only when it names one of those accounts. A reply from any of them, affiliates included, counts as the brand answering.

Which posts a brand is held to

A post can count in two ways: it tags the brand's main or support account and starts a conversation of its own, or it replies directly to a post from the brand's main, support or affiliate accounts.

Most posts that name a brand don't need anything from it. A ticker mention, a meme or a headline isn't a question, and counting those would turn a busy brand's score into a measure of its noise.

So each post is read by Vibewatch's reply scorer, the same one behind the reply queue Vibewatch gives its own customers. It decides whether the post needs a reply, why, and how relevant it is to the brand. A post counts only if the scorer's default setting, the middle of three, would put it in a team's reply queue as something to answer.

Praise counts only when the scorer rates it 8 or higher out of 10, the band the reply queue calls "Needs a reply". Up there it's usually a customer's own thanks, and a brand answering that is what the rankings measure. Below 8 it's mostly shout-outs, partner announcements and projects promoting themselves, which no brand answers one by one, so we leave it out. Likes are invisible to us, so a brand that only likes the praise it gets reads as not answering it. For a ranking of replies, that's intended.

The scorer makes mistakes in both directions. It will flag some posts a reasonable team would skip, and miss some that a team answered. Every brand is read by the same scorer, but its mistakes aren't spread evenly: a brand whose community posts in another language, or about unusual topics, can be read less accurately than the rest.

What counts as an answer

A post that starts a conversation counts as answered when one of the brand's accounts replied in public within 72 hours: directly to the post, or anywhere in the thread it started. A reply in the thread is credited even if it answered someone else in that thread, so the score measures whether the brand showed up, not whether the reply solved anything.

A reply under one of the brand's own posts counts as answered only when the brand replied directly to it. The thread is the brand's own, so a brand reply elsewhere in it is answering someone else.

For those replies we don't read the brand's whole thread. We search the brand's own posts for a reply to the person, and the post counts only where one complete search of every brand account covered its full 72 hours. Finding nothing then means the brand posted nothing that search could see. We checked the search against 20 posts whose whole thread we did read and that had a direct reply from the brand, and it found all 20. That check is small and flatters the search a little, because a reply only the search saw counts as found. A reply it misses reads as no answer, so if the search errs, these posts understate a brand.

Replies to anyone else's post are left out. A brand can be seen replying to a post deep in someone else's thread, but its silence there can't be seen, so those posts would only ever add answers. They're left out of both sides. So is a reply to a brand's reply inside a customer's thread, where the brand's next post there could be answering either person.

The window and the score

Each brand is measured once, over 30 posting days, with the exact dates printed on its report. Collection runs four days behind the calendar: a brand gets 72 hours to reply before a post can count as unanswered, and each day is read twice, a day apart, to catch the replies the first read missed. A window closes before its report comes out, sometimes by a few weeks, so the dates on the report are the ones that count.

A brand's score is the share of its counted posts that were answered. Each score comes with an approximate 95% range. That range ignores the way posts cluster around incidents and launches, so it's narrower than the truth. When two brands' ranges overlap, the data can't tell them apart, whatever order they're printed in.

A brand is ranked only when three things hold. Its sample is the equivalent of at least 20 counted posts, every one of its counted posts was scored, and we collected it for the whole window. A brand short on any of them is shown unranked. Brands are measured over different 30 days, so a published score is finished and does not move afterwards. Every score on the board comes from the same frozen version of the scorer, which is what lets one brand's number be read against another's.

The posts each report shows

A missed reply is easy to read as a nice-to-have, so each report shows what went unanswered. It sorts a sample of the highest-scored missed posts by kind, such as support issues, questions and feedback.

Where we can, the report then embeds three to five unanswered posts from X: the most-engaged support issues and questions. Under each we give its kind, its likes and replies, whether anyone in the community answered it when we read its thread, and that we found no public reply from the brand within 72 hours. That last line is what the collection saw. As the next section explains, the collection can miss replies, on X as well as off it.

A post is shown only if the scorer flagged it and the collection saw it go unanswered. We never go looking for posts to show. We don't quote, screenshot or tag the person who wrote one: the embed shows the post itself, and it disappears if they delete it. A post that shows personal details or personal hardship is skipped, because the report is about the brand's silence.

What the rankings can't see

  • Private answers. DMs, support tickets, Discord and email are invisible. A brand that moves its support into DMs will score lower than the help it gives.
  • Late or deleted replies. A reply after 72 hours, or one the brand deleted, counts as no reply.
  • Every post. Mentions are read in six-hour blocks with a cap on each. A block that hits the cap is left out, and on a heavy day only some blocks are read, with those posts weighted up to stand for the rest. Rates use those weights. Post counts are the posts we saw. A reply under a brand's own post counts only when a complete search of the brand's accounts covered its whole 72 hours.
  • Every thread. A post with no replies is easy to label. A long or busy thread sometimes can't be read in full, so the conversation starters we label lean toward quieter threads. Some brand replies in a thread we couldn't fully read are recovered from the brand's own posts.
  • What happened that week. An outage or a launch can bury a support team for a day. Every score is stored with the day that held the most of the brand's posts and the rate without that day, so a report can say when one day decided it.

If you think your score is wrong

Reply to the report on X, or message @Vibewatch_io, with the post and the reply we missed, or an account of yours we should be counting. We'll check it against what we collected. If we got it wrong, we fix it and the next report says what changed.

Related articles

One weekly report on your community, with citations.

7-day free trial, no credit card required.

Start free trial