Twenty seconds of what this looks like: the contract, the word landing in the transcript, and the price catching up afterwards.
What a mention market actually is
A mention market is a contract that settles on whether a specific word or phrase is spoken during a defined event. Someone picks a term, someone picks a speaker, and the market prices the odds of that term coming out of that mouth. Shares trade between roughly a cent and ninety-nine cents, and the correct side pays a dollar.
Both major venues list them. On Kalshi the catalogue runs across NFL broadcasts, corporate earnings calls, political speeches, and FOMC press conferences; Polymarket leans into social media, viral moments, and broader broadcast rules. The NFL category alone did roughly $47m of volume across 64 games in the 2025-26 season, and Super Bowl LX listed 34 separate terms on Kalshi against 27 on Polymarket.
The structure is what makes them unusual to trade. Nothing about the outcome exists until someone says it out loud. There is no filing, no data release, no scheduled print. The resolving event is audio, and it happens once, in real time, in front of everybody.
The edge is reaction time, not opinion
Pre-event research gets you a price, not an edge. Counting how often an announcer says a phrase across ten seasons, or how a CEO has used a term on the last twelve calls, is genuinely useful work - and it is also the work everyone else in that market did, which is why the pre-event price already reflects it.
What is not priced is the interval between the word being spoken and the market finding out. In that window the contract is worth a dollar and is still trading at twenty cents, because the information exists in the air and nowhere else yet. Everyone is looking at the same feed. The question is what stands between the audio and your order.
Usually one of three things, all slow:
- A person listening. Works until the event runs three hours, or two events overlap, or the phrase lands in the ninety seconds you spent reading something else. Attention is the resource that fails first.
- A news article. Reliable, well after the fact. By the time a sentence is quoted in a story, the move happened.
- The chat, or a Discord ping. Fast when someone else is paying attention, silent when they aren't, and you have no way of knowing which situation you're in until it's over.
The alternative is a machine that never stops listening and never gets bored: transcribe the broadcast as it airs and fire on the exact phrases the contract settles on. Lines land within seconds of being spoken, and the alert reaches you whether or not you were looking at the stream.
Read the resolution rules before you tag anything
This is where the most money is lost, and it has nothing to do with speed. Two venues can list what looks like the same contract on the same broadcast and settle it differently. Four details decide it, and they are worth reading in full every time:
- The exact phrase. Whether root words, plurals, and variants count, or only the literal string.
- Who has to say it. Kalshi's NFL mention markets settle on the designated play-by-play and colour commentators only - for Super Bowl LX, that meant Mike Tirico and Cris Collinsworth and nobody else. Polymarket resolves if anyone on the broadcast says it, sideline reporters and field mics included.
- Where it has to appear. Spoken audio, closed captions, or the official written transcript. These do not always agree.
- The source of truth. An official transcript, a named list of media outlets, or the broadcast itself.
That wider net is why identical terms often price higher on Polymarket than on Kalshi, which is a spread you can only see if you read both rulebooks. It is also how a market resolves against what you plainly heard. In February 2026, a Sanders rally market that traded $3.6m of volume resolved No despite video of the term being used, because Kalshi's rules required verification from named outlets that had not covered the closed event.
The practical consequence for your setup: tag what the contract says, not what you mean. If the market settles on the literal phrase, tag the literal phrase. If it counts anyone on the broadcast, your transcript needs to cover the whole feed, not one microphone.
The setup, end to end
Five steps, and four of them happen before the event begins. That is deliberate - the only thing you should be doing while the event airs is deciding.
1. Pull the phrases off the market page
Copy the terms word for word from the contract, and note the speaker and resolution source alongside each one. If you're trading the same event on both venues, keep the two lists separate. They are not the same trade.
2. Start the transcript before the event does
Paste the URL of the stream carrying the event into LiveScript and start the capture. Transcription runs from the moment you start, not from the start of the broadcast, so early beats punctual. YouTube and Twitch streams take a URL directly; for an earnings webcast or an investor-relations player with no pasteable link, the browser extension transcribes whatever audio is playing in the tab.
3. Tag every phrase you have a position in
Add each term as a keyword tag. Matching is whole-word and includes the common English plural, so tariff fires on "tariffs" and assist does not fire inside "assistant" - which matters when a false ping costs you a fill. Tag generously; an alert you ignore is cheap, and a phrase you didn't tag is invisible.
4. Stop watching
Put the stream in a background tab and do something else. When a tagged phrase is spoken the alert fires and links to the exact transcript line. This is the part that changes the workflow rather than speeding it up: you can cover three simultaneous events, or a six-hour hearing, without holding attention on any of them. The longer walkthrough is in monitoring live streams for spoken keywords.
5. Read the line, then act
The alert tells you where to look; the transcript tells you what was said. Confirm the words with your own eyes before you size up, because speech recognition is not perfect and a phrase near your tag is not your tag.
Events that start when they feel like it
Half the useful mention markets hang on a broadcast with no reliable start time: a rally that begins forty minutes late, a channel that goes live unannounced, a presser bolted onto the end of something else. Starting a capture manually means either watching for the stream to appear or missing the opening.
Channel drop alerts handle that: LiveScript watches a YouTube channel, emails you the moment it posts or goes live, and starts the transcript automatically. By the time you open the tab, the transcript is already running and the first ninety seconds are in it.
If you'd rather not be in the loop at all
Alerts still route through a human. If your matching logic already lives in code, the live WebSocket API pushes transcript lines to your program as they are spoken, so your own phrase matching runs against the text directly:
wss://livescript.live/api/v1/liveEndpoints, auth, and event shapes are in the API reference. For post-event work - checking whether a term actually appeared, or backfilling a finished video before you research the next contract - the MCP server puts a finished transcript in front of Claude or any other MCP client. Reach for the WebSocket while the event is airing and MCP once it's over.
What "fast" honestly means here
Three things are worth being precise about, because a mention market punishes vagueness.
The lag is seconds, and it cannot be zero. Audio has to reach the transcriber and be transcribed before words can appear. Anyone advertising zero-latency transcription is describing something that does not exist. The benchmark that matters is not zero - it's the twenty minutes to an article, or the indefinite delay of a human who looked away.
Your feed delay is the same as everyone else's - mostly. Broadcast and streaming delays affect every trader watching the same source, so they cancel out across the market. They do not cancel out between sources: a low-latency feed and a thirty-second-behind restream are different events as far as your fill is concerned. Transcribe the fastest feed you legitimately have.
Recognition errors cluster on exactly the wrong words. Proper nouns, tickers, unusual surnames, and crosstalk are the hardest tokens for speech recognition - and they're disproportionately what mention markets settle on. This is the reason the product pairs an alert with a readable transcript rather than replacing one with the other. The alert says where to look; you confirm what was said.
The short version
Mention markets are decided by audio, and audio is the one input that reaches every trader at the same moment. The difference between the traders is what happens in the seconds after the word is said: whether it reaches a person who was listening, an article twenty minutes later, or a tag that fires the instant it hits the transcript.
Read the rulebook first, tag the literal phrases, start the capture early, and let the alert find you. Paste a live URL into LiveScript before the next presser and you'll be reading the words while everyone else is still waiting to hear about them.
Nothing here is trading advice. LiveScript transcribes broadcasts and tells you when a phrase was said; what you do with that is yours.