Strategies·November 17, 2026·7 min read

Mention Markets on Polymarket: Trading What a Speaker Will Say

How to trade mention markets on Polymarket: counting base rates in past transcripts, mapping speech structure and timing, matching the exact phrase rule, hedging related phrases, and pricing viral moments.

mention marketstranscriptsevent tradingpolymarketresolution rules
Social media — illustrative image
Photo: Words as Pictures / StockSnap · CC0Source

Mention markets ask a narrow question: will this person say this word? The contract pays $1 if the phrase appears in the defined transcript and $0 if it does not. It looks like a novelty market. It is really a text search with a deadline, and the traders who do well treat it like one.

The edge is not in guessing the mood of a speech. It is in knowing the base rate, the structure of the event, and the exact matching rule better than the person on the other side of a thin book.

The contract is a text search

Start with the literal string. A well-written rule names the speaker, the event, the time window, the transcript source, and the phrase. It also answers the questions that decide close cases: does a plural count, does a stem count, must the words be contiguous, does a question from a reporter count, and what happens if the event is cancelled or the speaker is replaced.

"Will mention" and "will say" are not the same contract. A mention can appear on a graphic, in a question, or in a quote read aloud. A rule that requires the named speaker to utter the phrase is narrower, and usually cheaper. Read the resolution source too. If the market resolves on the official transcript published hours later, a live caption that mishears a word is not evidence.

Write the matching rule as a sentence before you trade: the exact words, by this speaker, in this venue, between these times, according to this transcript. If you cannot write it, you do not know what you own.

Base rates from past transcripts

The single most useful input is a corpus. Collect official transcripts for the same speaker and the same kind of event, then count how often the phrase appeared. Record the denominator, not just the hits.

Suppose a phrase appears in 9 of 48 comparable speeches, a base rate of 18.75%. That number is a starting point, not a forecast. Split it by section: 2 of 48 in prepared remarks and 7 of 48 in questions. Split it by recency: 4 of the last 8 versus 5 of the first 40. A phrase tied to a current news cycle should be weighted toward the recent regime.

Novel phrases have no base rate. When a market asks whether a speaker will use a phrase that has never appeared, the honest prior is low but uncertain, and the distribution is wide. In that case, estimate from analogous phrases and from the topic's salience, then size for the width of the range rather than the point estimate.

Speech structure and timing

Speeches are not uniform. There is an opening, a set of thank-yous, the prepared argument, and often questions. Mention probabilities cluster: names of local officials at the start, policy terms in the middle, and unscripted phrases in Q&A.

The timing matters as much as the level. If a phrase usually appears in the Q&A, a contract at 0.30 with ten minutes left in the prepared remarks is not mispriced at 0.30. The price should decay as the remaining opportunities shrink. Traders routinely under-update this hazard rate, which is one of the few clean edges in the category.

There is also an information asymmetry before the event. Prepared remarks are often distributed to reporters under embargo, and official statements are released at a scheduled minute. If the text is public before delivery, the market should already be near 0 or 1, and buying after the release is buying stale news.

Market mechanics and costs

Mention books are thin. A typical quote might be 14 cents bid, 20 cents offered, with a few thousand shares of depth. Buying 2,000 shares at 0.20 costs $400. At the politics taker rate of 0.04, the fee is 2,000 x 0.04 x 0.20 x 0.80, or $12.80, which is 3.2% of the stake. Break-even probability rises from 20% to 20.64%.

Holding to resolution removes the exit cost. Trading in and out does not. Selling the same position at 0.16 costs 2,000 x 0.04 x 0.16 x 0.84, or $10.75. Round-trip fees are $23.55, and the four-cent gap between the 0.20 purchase and the 0.16 exit costs another $80. The round trip consumes about 26% of the $400 position before any probability error. That is why patient limit orders matter, and why most mention trades should be either maker trades or hold-to-resolution trades.

Hedging across related phrases

Phrase markets on the same speech are correlated, but usually not exclusive. A speaker can mention inflation and also mention recession. Buying both Yes contracts is a bet on two related events, not a hedge.

The clean hedge exists only when the rules make outcomes mutually exclusive. "Will say the word tariffs" and "will not say the word tariffs" should sum to one, minus costs. If they trade at 0.62 and 0.45, the pair costs 1.07 and there is no free money after fees and spread. If they trade at 0.55 and 0.38, the pair costs 0.93, and buying both locks in about 7 cents before fees, provided both resolve on the same transcript.

Baskets are the more common structure. A market may ask whether the speaker mentions any of three names, while three separate markets price each name. If the sum of the individual prices is above the basket price, and the rules match, buy the basket and sell the individuals. If the rules use different transcripts, you have a correlation trade, not an arbitrage.

Polymarket's negative-risk mechanism helps when outcomes are genuinely exclusive, because one No share converts into Yes shares in the other outcomes. It does not apply to a list of phrases that can all be said. Check the mechanism before assuming a hedge.

Why viral moments are hard to price

The moments that produce the biggest moves are the ones with the least history. A novel insult, a slogan, or an unexpected admission has a base rate near zero and a payoff driven by one decision under pressure. There is no stable frequency to anchor the price, so the market trades on attention.

Attention is a bad probability estimate. Social posts and clips push prices in thin books, and the people buying often conflate "the topic will come up" with "the exact phrase will be said". The rules rarely reward the topic. They reward the string.

Resolution risk is highest here too. A phrase that is almost said invites dispute, and a disputed market can take days to settle. A contract bought at 0.70 that resolves 50/50 costs 20 cents per share even though the speaker clearly said something close.

How to execute

  1. Build the corpus first. Store official transcripts by date, speaker and venue.
  2. Encode the rule as a matcher: exact phrase, variants, speaker, window.
  3. Count occurrences per event, and split by section and recency.
  4. Convert the base rate into a time path, not a single number.
  5. Check the book depth and the fee before sizing.
  6. Prefer limit orders, and assume the exit costs the full spread.
  7. Write down the resolution source and the publication time.

Risk controls

The main risks are rule risk, transcript risk, event risk and liquidity risk. Rule risk comes from words like "mention", "discuss" and "address". Transcript risk comes from automated captions and later corrections. Event risk includes cancellations and speaker substitutions. Liquidity risk means the price you see is not the price you can exit at.

Cap exposure per speech, not per phrase. Ten markets on one event are one bet on the speaker's mood and one bet on the same transcript. A practical limit is 1% to 2% of bankroll per speech across all phrases, with no single phrase above 0.5%.

Checklist

  • The rule names the speaker, window, phrase and transcript source.
  • You have counted the phrase in at least 20 comparable past events.
  • You know which section of the speech produces the phrase.
  • The price reflects the time remaining, not just the terminal base rate.
  • Round-trip cost is under a third of the expected edge.
  • Related phrase markets are treated as one correlated position.
  • The resolution transcript is published by a source you can check live.
  • You have a written exit plan for a disputed resolution.

Mention markets reward preparation more than prediction. The speaker is going to give the same kind of speech as last time. The transcript is going to be published by the same office. The only variable is the phrase, and the phrase can be counted.

Sources

  1. 1.Resolution— Polymarket Documentation
  2. 2.Markets & Events— Polymarket Documentation
  3. 3.Prices & Orderbook— Polymarket Documentation
  4. 4.The American Presidency Project: Documents and Transcripts— University of California, Santa Barbara
  5. 5.Debate Transcripts— Commission on Presidential Debates
  6. 6.Speeches and Testimony Archive— Federal Reserve Board

More from the journal

All articles