Reading earnings calls is slow. A single transcript runs tens of thousands of words, and a busy quarter means dozens of them stacking up faster than anyone can read. That is exactly the kind of work large language models are good at, once they can reach the source text.

This is a list of 10 concrete, copy-pasteable Claude earnings call prompts. Each one is built to work with the earningscalls MCP connector, which gives Claude live access to 246,000+ earnings call transcripts across 12,000+ companies, 70 countries, 170+ exchanges, and all 11 GICS sectors, from 2020 to today. When you paste one of these prompts, Claude reaches into that dataset through tools like search_transcripts, get_transcript, and get_speaker_segments and reasons over the real words executives and analysts actually said.

Before You Start: Connect the MCP Server

These prompts only work once Claude can read the transcripts. Website reading on earningscalls.dev is free, but to analyze earnings calls with Claude directly inside your chat, you need the connector.

  1. Go to your dashboard → Connectors → Generate. You will get a personal URL like https://earningscalls.dev/u/mct_xxxx/mcp.
  2. In Claude Desktop or claude.ai, open Settings → Connectors → Add custom connector and paste that URL.
  3. In Claude Code, run claude mcp add --transport http earningscalls https://earningscalls.dev/u/mct_xxxx/mcp.

Full setup instructions live on the MCP page and in the docs, and there is a step-by-step walkthrough in How to connect earnings call transcripts to Claude with MCP.

One tip that runs through every prompt below: when a query is ambiguous, naming the tool nudges Claude to use it instead of guessing. Starting with "Using the earningscalls tools, ..." is the single most reliable way to keep answers grounded in real transcript text.

1. The Single-Call Summary

Using the earningscalls tools, pull the most recent earnings call for NVIDIA and give me a one-page summary: headline results, the three most important things management said, guidance, and any change in tone versus what analysts expected.

This is the workhorse prompt. Claude finds the latest call, retrieves the transcript, and condenses it into something you can read in a minute. It typically invokes get_latest_call_for_ticker to locate the call and get_transcript (in summary format) to read it. Use it the morning after any print to get oriented before you dig deeper.

2. CEO Tone vs. CFO Tone

Using the earningscalls tools, compare how the CEO and the CFO spoke on Microsoft's last earnings call. Where is the CEO optimistic while the CFO is cautious? Quote the exact lines that show the gap.

CEOs sell the story; CFOs guard the numbers. When their framing diverges, that gap is often where the real signal hides. Claude uses get_speaker_segments with a role filter to separate executive turns by speaker, then contrasts their language side by side. It is a fast way to catch a management team that is more hopeful on stage than the financials justify.

3. Prepared Remarks vs. Q&A

Using the earningscalls tools, split Tesla's latest call into prepared remarks and the Q&A section. Summarize each separately, and flag anything management admitted under questioning that was not mentioned in the scripted portion.

Prepared remarks are rehearsed and polished. The Q&A is where executives improvise, hedge, and occasionally reveal what they would rather not. Claude reads the full transcript, uses speaker roles and ordering to identify where the operator hands off to analysts, and reports the two halves distinctly. The interesting material almost always sits in the difference.

4. Cross-Quarter Comparison

Using the earningscalls tools, compare Amazon's last four earnings calls. How has management's messaging on margins and AWS growth evolved quarter over quarter? Show me the trajectory, not just the latest snapshot.

A single call is a data point; four calls are a trend. This prompt drives Claude to pull the ticker's recent archive with list_calls_by_ticker, retrieve several transcripts, and track how a specific theme moves over time. It surfaces slow narrative drift that is invisible when you only read the current quarter.

5. Cross-Company Theme Scan

Using the earningscalls tools, search the last two quarters of transcripts for mentions of "AI capex" across large-cap tech. Which companies are raising spend, which are holding it flat, and what reasons do they give?

This is where the MCP dataset earns its keep. Instead of reading twenty calls, Claude runs search_transcripts for the theme across the whole corpus and returns the matching passages with company and speaker attribution. Swap "AI capex" for "tariffs," "guidance raised," "pricing power," or any phrase you care about to build an instant thematic map of the market.

6. Analyst-Question Analysis

Using the earningscalls tools, list every analyst question from Apple's most recent call, group them by topic, and tell me which subjects the analysts pushed hardest on. Where did management give a direct answer versus a non-answer?

What analysts choose to ask reveals what the smart money is worried about. Claude pulls the analyst-role speaker segments, clusters the questions, and evaluates the quality of each response. Repeated questioning on one topic, or a string of evasive answers, is a signal in itself. There is a deeper treatment of this technique in How analysts read earnings calls 10x faster with Claude and MCP.

7. Guidance-Change Tracking

Using the earningscalls tools, find every place across Meta's last three calls where management gave forward guidance on revenue, margins, or capex. Show me exactly how each figure or range changed from one quarter to the next.

Guidance revisions move stocks more than the reported quarter does. This prompt has Claude search across several calls specifically for forward-looking statements, then diff the numbers and language so you can see raises, cuts, and quiet withdrawals at a glance. It turns a tedious cross-reference into one paragraph.

8. Red-Flag and Hedging Language

Using the earningscalls tools, scan Boeing's latest earnings call for hedging and evasive language: phrases like "we're comfortable with," "as we've said," "puts and takes," or deflected questions. Quote them and explain what each one might be softening.

Executives rarely lie outright, but they hedge, deflect, and bury bad news in qualifiers. Claude reads the transcript and flags the linguistic tells that human readers skim past when they are tired. Pair the output with the numbers to decide whether the hedging is routine caution or a genuine warning sign.

9. Sector-Wide Sweep

Using the earningscalls tools, look across the most recent quarter of earnings calls in the semiconductor industry. What is the shared story on inventory and demand, and which companies are the outliers going against the consensus?

Sometimes the question is not about one company but about an entire sector's mood. Claude combines list_sectors / list_industries to scope the universe with search_transcripts to read across it, then synthesizes the consensus and highlights who is breaking from it. The outliers are usually the most interesting names to research next. For more on this pattern, see Screen earnings calls across the market with MCP tools.

10. What Changed Since Last Quarter

Using the earningscalls tools, compare Salesforce's two most recent earnings calls and give me a short "diff": what management stopped talking about, what is new this quarter, and any metric or priority that quietly disappeared.

The most valuable insight is often what went missing. A KPI that vanishes from the script, a product line that stops getting airtime, a confident phrase replaced by a vaguer one, these are all signals. Claude retrieves both transcripts and reports the delta in plain language, so you catch omissions that a straight read would never reveal.

Getting the Most Out of These Prompts

A few habits make these analyze earnings calls with Claude workflows dramatically better:

Because the dataset covers 11M+ speaker segments with roles for executives, analysts, operators, attendees, and shareholders, you can slice a call almost any way you want, by who spoke, by section, by theme, or across an entire sector, without leaving the chat.

Start Analyzing

These prompts turn Claude into a tireless earnings-call analyst that reads faster than any human and never skips the Q&A. The only prerequisite is the connector.

Connect the MCP server from your dashboard, then paste any prompt above and watch Claude reason over the real transcripts. Pro is $24.99/mo (5,000 requests/mo) and Ultra is $39.99/mo (25,000 requests/mo); see the full breakdown on the pricing page. Reading transcripts on the website is always free, so you can explore the data before you ever wire up an API key.