Skip to main content
← Back to Docs
✨AI Tools

AI Market Summaries

Auto-generated briefs of the biggest mispricings & moves

AI Market Summaries turn the firehose of Polymarket and Kalshi activity into a short, readable briefing you can skim in a couple of minutes. Instead of manually paging through the EV Scanner, refreshing order books, and trying to spot what changed since this morning, you open **/dashboard/summaries** and get a ranked stack of cards that each tell you what a market is doing right now, why it might matter, and what to watch next.

Think of it as the "what happened" layer on top of the raw data. The scanner shows you every number. Summaries tell you which numbers are worth your attention today.

## What It Does

Predite runs a scheduled job several times a day that pulls the most active markets across the platforms we track, then generates a natural-language brief for each one. The output is a list of market cards, sorted so the highest-activity, highest-volume markets sit at the top.

Each card condenses four things into a glance:

- **A plain-English summary** — 2 to 3 sentences describing what the current price implies, whether the market is busy or quiet, and any notable pattern (extreme pricing, a volume spike, an approaching resolution).

  • •**A sentiment tag** — `bullish` (green), `bearish` (red), or `neutral` (white). This reflects how the market is currently leaning on the YES side, not a recommendation to buy.
  • •**Key insights** — up to four bullet points calling out the specific, actionable thing: a possible mispricing, whale-scale volume, thin liquidity, or a market resolving soon.
  • •**Market vitals** — current YES price with a direction arrow (▲ up, ▼ down, — stable), 24h volume, category, and an activity level badge (🔥 high, 📊 medium, 💤 low).

When an Anthropic API key is configured on your workspace, summaries are written by Claude and the header reads **Powered by Claude**. If the AI service is briefly unavailable, Predite falls back to a deterministic heuristic engine so the page never goes blank — you'll see **Heuristic analysis** in the header instead. The heuristic version is rule-based (price thresholds, volume tiers, days-to-resolution) and still surfaces the same categories of insight, just in more templated language.

### What each summary highlights

The whole point is signal over noise, so the engine is tuned to flag the things traders actually act on:

- **Biggest mispricings.** Markets sitting at extreme probabilities (above ~90% or below ~10%) get flagged for possible favorite–longshot bias. Markets hovering near 50/50 get flagged as maximum-uncertainty, sharp-move candidates. These are the two ends where EV most often hides.

  • •**Whale moves and unusual volume.** When 24h volume is high (seven figures and up), the brief notes likely institutional or whale interest and treats it as a sign that new information may be driving price. Quiet, low-volume markets get the opposite call-out — "undertracked, potential opportunity for informed traders."
  • •**Resolving markets.** Markets within a week of resolution are tagged as time-sensitive, and anything inside ~3 days gets an explicit "expect increased volatility" note. Resolution windows are where pricing converges fast and stale positions get punished.
  • •**Liquidity risk.** Thin order books get a warning that large orders may move the price, so you size accordingly.

## How to Use It, Step by Step

1. **Open the page.** Go to **/dashboard/summaries** from the sidebar. On load, Predite fetches the latest batch of summaries (15 markets by default). The header shows the data source (Claude or heuristic) and the timestamp of the most recent generation. 2. **Read the sentiment overview.** At the top, three tiles show how many of the current summaries are bullish, bearish, and neutral. This is your instant market-wide mood read before you dig into individual cards. 3. **Filter to what you care about.** Click any sentiment tile to filter the list to just those markets (click again to clear). Use the activity filter row (**ALL / 🔥 high / 📊 medium / 💤 low**) to narrow further. Filtering on "high activity" alone is the fastest way to see where the action is concentrated right now. 4. **Scan the cards top-down.** They're already sorted by activity then volume, so the most important markets are first. For each card, read the one-line question, glance at the price + arrow + volume, then read the AI summary and the **KEY INSIGHTS** bullets. 5. **Act on the standouts.** When a card flags something worth a position — a mispricing, a whale-driven move, a resolving market — that's your cue to jump into the deeper tools (see [How It Connects to Other Features](#how-it-connects-to-other-features) below). 6. **Refresh on demand.** Hit the **⟳ Refresh** button to pull a fresh batch immediately. This is useful right after a big news event, but note the rate limit below — refreshes are not free.

### A practical daily routine

The intended workflow is to use Summaries as your morning (and midday) triage instead of doom-scrolling:

1. Open Summaries first thing. Read the three sentiment tiles for the overall lean. 2. Filter to **🔥 high activity** and read those 3–6 cards carefully. These are where money and information are moving. 3. Note any **resolving in N days** or **extreme probability** flags — those are your time-sensitive and mispricing candidates. 4. For each candidate, open the market detail or the EV Scanner to confirm the edge with real numbers. 5. Come back midday. Because summaries regenerate several times a day, the second read will reflect new price action and volume without you having to track it manually.

That loop takes a few minutes and replaces an hour of refreshing tabs.

## How the Generation Schedule Works

Summaries are produced by a server-side cron job, not generated live on every page view. The market-scanning job runs on a fixed schedule (roughly every four hours, so multiple refreshes across the trading day), pulling the most active markets and writing a brief for each. Because it's cron-driven:

- The data you see is **as of the last run** — the timestamp in the header tells you exactly when. If it says the batch is a couple of hours old, that's expected; a new one is coming.

  • •Generation runs server-side with proper authorization, so it doesn't count against your personal rate limit.
  • •The list is **ranked, not exhaustive.** A single run summarizes the top markets by activity (up to 20), not every market on every platform. If a market you care about isn't in the list, it's almost certainly because it's quiet right now — which is itself a signal.

When you click **Refresh**, you're requesting a fresh on-demand generation tied to your account rather than waiting for the next cron cycle.

## Plan Requirements

AI Market Summaries are part of the **Pro plan ($59/mo)** and are also included in the **Bot plan ($99/mo)**. They are not available on Starter ($29/mo).

If you're on Starter, the page shows an upgrade screen listing what Pro unlocks (unlimited real-time signals, all platforms, the WebSocket feed, whale tracking, arbitrage, the Kelly Calculator, backtesting, and **AI Chat + AI Summaries**) with a one-click path to upgrade. Summaries are bundled with AI Chat under the same Pro entitlement, so unlocking one unlocks both.

Plan access is enforced both in the UI and at the API layer, so there's no way to pull summary data without the right tier — and no reason to, since the upgrade also lights up the tools you'd use to act on the summaries.

## Tips

- **Use sentiment as context, not a trade signal.** A `bullish` tag means the crowd is leaning YES, often *because* the price is already high. The edge is frequently in disagreeing with a confident crowd, not following it. Pair the tag with the EV and probability tools before sizing anything.

  • •**Treat "low activity" cards as a watchlist, not noise.** The engine explicitly flags quiet markets as potentially undertracked. If you have an edge in a niche category, a low-volume card with a mispricing flag can be more valuable than a crowded high-volume one.
  • •**Lead with the KEY INSIGHTS line.** The prose summary sets context, but the bulleted insights are where the specific, actionable callouts live (extreme probability, thin liquidity, resolving soon). Skim those first across all cards, then read the full summary only for the ones that hook you.
  • •**Cross-check the direction arrow against the EV Scanner.** The arrow reflects which side the price is leaning, not momentum over a fixed window. If a card shows ▲ but your model says the true probability is lower, that's exactly the kind of gap worth opening the scanner to confirm.
  • •**Refresh after catalysts, not out of habit.** A scheduled debate, an economic print, or a major headline is a great reason to force a refresh. Idle re-refreshing just burns your hourly quota.

## Gotchas

- **The page is a snapshot, not a live ticker.** Prices and volume reflect the last generation run. For live, tick-by-tick data, use the real-time feed and market detail pages. Always sanity-check the timestamp before trading off a summary, especially around fast-moving events.

  • •**Manual refresh is rate-limited.** On-demand refreshes are capped at **10 per hour per account** because each one triggers an AI generation pass. If you hit the cap you'll get a "rate limit exceeded" message and need to wait before refreshing again. The scheduled batches keep flowing regardless — the limit only applies to your manual pulls.
  • •**Sentiment is about price lean, not your P&L.** A `bearish` tag on a market you hold YES in doesn't mean "you're wrong," it means the crowd is pricing NO. Read it as market positioning, not a verdict on your thesis.
  • •**Heuristic mode is more conservative.** If the header reads "Heuristic analysis," the summaries are rule-based rather than AI-written. They're still accurate on the structural facts (price, volume, liquidity, days to resolution) but won't catch nuanced, context-specific patterns the way Claude does. This is a fallback state, not the default.
  • •**Coverage is the top markets, not all of them.** Don't assume a market is fine just because it isn't flagged. Absence from the list means low activity, not "nothing to see here."

## How It Connects to Other Features

Summaries are designed as the **entry point** of your workflow, handing you off to the tools that do the deep work:

- **EV Scanner** — When a card flags a possible mispricing, open the scanner to see the exact expected-value math and confirm the edge across platforms.

  • •**AI Probability** — Use Predite's modeled fair-value estimate to test whether the crowd lean in a summary is actually wrong, and by how much.
  • •**Whale Tracker** — When a brief notes whale-scale volume, jump to the tracker to see who's moving size and whether you want to follow.
  • •**Kelly Calculator** — Once you've confirmed an edge from a summary, size the position properly instead of guessing.

Summaries tell you *where* to look and *why*. The tools below tell you *how much* and *whether to pull the trigger*.

## Related Docs

- [EV Scanner](/docs/ev-scanner)

  • •[AI Probability](/docs/ai-probability)
  • •[Whale Tracker](/docs/whale-tracker)
  • •[Reading Signals](/docs/reading-signals)
← All Documentation
AI Market Summaries | Predite