Most earnings call summaries miss the only thing that matters: what changed since last quarter. Educational only. Not investment advice.
All briefs
Wall Street Prompt · LinkedIn one-page
How To Analyze An Earnings Call With Claude
LinkedIn one-page
Most earnings call summaries miss the only thing that matters: what changed since last quarter.
LinkedIn1080 × 1350
References & validationExact post missing
Reference creator resolved, but the exact source post was not captured. Capture the specific post (not just the profile) before treating as validated.
Reference creator: Rachel Woods / Riley Brown
Post content
Dense one-page content — sections, tables and frameworks
The summary trap
A generic summary tells you what was said. It does not tell you what moved. The signal in an earnings call lives in the deltas, not the recap.
- Summaries flatten the call into bullet points everyone already read in the press release.
- They repeat guidance without checking it against last quarter's guidance.
- They quote management optimism and skip the hedging in the same paragraph.
- They miss the gap between the prepared remarks and the Q&A answers.
Why it matters: price reacts to revision, not repetition. If your read of the call is the same as the headline, you have no edge. The work is comparison, not comprehension.
Four things to actually extract
Point Claude at the transcript and ask it to pull these four buckets. Each one maps to a concrete question an analyst would ask.
| What to extract | The question it answers | Where it hides |
|---|---|---|
| Changes vs last quarter | What is management saying now that they were not saying 90 days ago? | Guidance ranges, margin commentary, capex tone, segment priority shifts |
| Contradictions | Where do the prepared remarks and the Q&A answers disagree? | Scripted optimism up top, qualified answers under analyst pressure |
| Risk language | What words are doing defensive work? | Macro, headwinds, normalize, prudent, monitoring, second half weighted |
| Non-answers | Which direct questions got deflected? | Repeated questions, pivots to long-term framing, declined to quantify |
Tip: the most useful output is not a paragraph. It is a list of items with the exact quote attached so you can verify the language yourself.
The two-transcript workflow
The single most valuable move is feeding Claude both the current and prior quarter transcripts and asking what changed. Run it in stages so each pass is verifiable.
| Step | Action | Output you keep |
|---|---|---|
| 1. Load | Paste current + prior quarter transcripts (and the press release) | Source set with dates labeled |
| 2. Diff guidance | Ask: how did stated guidance and tone shift quarter over quarter? | Side-by-side delta table with quotes |
| 3. Split the call | Ask Claude to separate prepared remarks from Q&A and compare confidence | Contradiction list, scripted vs unscripted |
| 4. Flag risk | Ask it to surface every hedge word and what it is qualifying | Risk-language inventory with context |
| 5. Pressure test | Ask: what would a skeptical analyst push back on here? | Open questions for your own IC memo |
Key: never let one prompt do all five. Staged passes keep each answer short enough to check against the transcript line by line.
Prompt scaffolding that works
Generic prompts get generic summaries. Constrain the task and force citation. Three patterns that hold up:
- Role + constraint: 'You are reading two earnings transcripts. Do not summarize. List only what changed between them, with the exact quote and quarter for each.'
- Force the quote: 'Every claim must include the verbatim sentence it came from. If you cannot quote it, do not include it.'
- Separate the sources: 'Tag each finding as PREPARED REMARKS or Q&A so I can see where management is scripted versus reacting.'
- Skeptic pass: 'Now argue the bear case using only language from this call. What did they avoid saying?'
Heads up: Claude can paraphrase confidently. The verbatim-quote rule is your hallucination check. No quote, no claim.
Where the workflow breaks
Know the failure modes before you trust the output. Each has a discipline that contains it.
| Common mistake | Failure mode | Disciplined fix |
|---|---|---|
| Asking for a summary | You get the press release back in different words | Ask only for changes, contradictions, and risk language |
| One mega-prompt | Long answer, unverifiable, blends sources | Stage the passes; one question each |
| Trusting paraphrase | Confident claim with no basis in the call | Require verbatim quotes for every point |
| Skipping the prior quarter | No baseline, so nothing reads as a change | Always load the comparison transcript |
| Treating output as a view | You outsource the judgment | AI builds the inventory. You decide what it means. |
Why it matters: the tool surfaces and organizes language at speed. It does not weigh it. Human judges. AI builds. The position sizing, the conviction, the call stays with you.
Caption
LinkedIn post copy
Visual design notes
- Near-black forest-green background (#0B1410 to #0F1A14 vertical gradient). ONE accent: a single bright analyst-green (#22C55E / #2DE38A) used only for the emphasized title word, table header rules, and the subtitle banner. Everything else is off-white (#E8EFE9) and muted gray (#7E8C82).
- Header: heavy condensed sans (Anton / Druk / Saira Condensed) all-caps headline, left-aligned, tight leading. 'ANALYZE' set in the green accent, the rest in off-white. Directly below, a thin teal-green banner bar holding the subtitle one-liner in lighter weight sans.
- Tables are the spine of the page: dense 2- and 3-column layouts with a thin green top rule per table, hairline row dividers in low-opacity gray, generous left padding, zebra rows at ~4% white. Column headers in small-caps green. Keep cell copy short and declarative so rows stay scannable.
- Section headings: small-caps green kicker number (01-05) + bold off-white heading. 'Why it matters' / 'Tip' / 'Key' / 'Heads up' notes set apart in a left-border callout (2px green vertical rule, italic-free, slightly muted text).
- Add one small diagram between Section 3 and 4 if vertical space allows: a simple two-box 'Prior Quarter -> Current Quarter' delta arrow with the word CHANGE on the arrow, reinforcing the two-transcript core idea. Minimal, line-based, green accent only.
- Strict left alignment throughout, no centered text except the headline block. Comfortable margins (96px sides on 1080 width). Footer locked to the bottom on a thin divider rule. No em dashes anywhere in rendered copy. Maintain high contrast for mobile LinkedIn feed legibility.
Production checklist
- ☐Build the 1080x1350 one-pager in the WSP template: forest/near-black gradient background, single green accent, condensed all-caps headline with 'ANALYZE' emphasized, teal subtitle banner bar.
- ☐Lay out all five sections left-aligned with small-caps green kicker numbers; render the three required tables (Four things to extract, Two-transcript workflow, Where the workflow breaks) as dense hairline-divided 2-3 column tables.
- ☐Set the two bulleted sections (summary trap, prompt scaffolding) and style every 'Why it matters / Tip / Key / Heads up' note as a left-border green callout. Proofread for zero em dashes and no buy/sell/hold language.
- ☐Add header and footer WSP branding: Dave Wang attribution line on a bottom divider rule, optional small WSP wordmark top-right. Insert the optional Prior->Current delta diagram if space allows.
- ☐QA pass at 100% and at mobile feed size for contrast and table legibility; confirm disclaimer present in caption.
- ☐Export final as PNG 1080x1350 for the LinkedIn feed and a matching PDF version for DM delivery of the prompt scaffold.
Follow Dave Wang · Wall Street Prompt · For more AI investing workflows · COMMENT & REPOST
CTA
Save this for next earnings season. Comment "DIFF" and I will send the two-transcript prompt scaffold.