Earnings call communications is the planning, scripting and follow-up that surrounds a public company's quarterly call. In 2026 the call has two audiences: the analysts on the line, and the AI engines that retrieve the published transcript to answer questions about the company for the next four quarters. Plan 21 days ahead, script for summarization, and publish the transcript in HTML within four hours.
Part of EPR's investor relations hub. By the EPR Editorial Team. Updated October 2026. Everything-PR and 5W AI Communications share common ownership; Everything-PR reports independently on the communications industry.
Why do earnings calls now shape AI answers?
Earnings calls shape AI answers because transcripts are published on crawlable sites within hours and engines retrieve them when investors ask about a company. Seeking Alpha, Quartr, AlphaSense, The Motley Fool and company IR pages all carry transcripts, so a CEO's exact words become the source text for answers about margins, guidance and strategy.
Analyst questions on a call also resemble the prompts investors type into an engine. "What is your margin outlook?" on the call is close to "What is the company's margin outlook?" in ChatGPT, and the answer given on the call is the one the engine can quote. An IR team that scripts only for the people on the line leaves the second audience unmanaged.
What do AI engines keep and drop from an earnings call?
AI engines keep guidance numbers, sentiment words such as strong or softening, forward-looking commitments and named competitors, and they drop caveats, conditional clauses and subordinate context. 5W AI Communications reports these patterns from an observational, sample-based audit of mid-cap transcripts and their summaries across ChatGPT Enterprise, Claude, Perplexity and Gemini. The audit is continuing and is not a controlled study.
Here is an illustrative example, not a real transcript. A CFO says: "We saw modest deceleration in our enterprise segment offset by continued strength in mid-market, and we remain confident in our full-year guidance range of $4.20 to $4.35." A summary can return: "The company reported strength in mid-market and reaffirmed full-year guidance of $4.20 to $4.35." The deceleration and the qualifier are gone.
That matters for legal protection. Forward-looking statement protection under Section 21E is built around cautionary language, so a summary that strips the caution leaves the protection on the filed document and none on the answer investors read. Counsel should review this gap with the IR team.
Two other patterns follow. Engines do not reliably separate forward-looking guidance from historical fact, so guidance from Q2 can be returned in Q4 as a current commitment. And when no clean source exists, engines have been observed to assume a guidance reaffirmation or to carry a prior-quarter framing past a strategy change. Track both in the scorecard below.
Why it works. A language model compresses a transcript by keeping the passages that answer the question most directly, and short declarative sentences with numbers answer directly while hedged compound sentences do not. Repeating the same line in the release, script, filing and IR page gives the model several matching passages to choose from. EPR has not published controlled test results for this; treat it as practitioner guidance until measured.
What does the 21-day pre-call calendar include?
The 21-day pre-call calendar covers narrative, guidance wording, rehearsal and distribution in three stages. The legal quiet period limits what management says publicly, but it does not stop the IR team from preparing.
Days 21 to 14: what gets locked?
The IR team locks one organizing thesis that connects results to strategy, because "we beat estimates" is a data point and not a narrative. It then writes the guidance language at the word level. "We are raising full-year guidance to reflect the demand we see in the pipeline" gives an engine more to quote than "we are confident in our ability to deliver."
The team also builds a Q&A book with answers to the 20 most likely analyst questions, each with the data point, the bridge to the thesis and the wording the CEO will use. The press release is drafted at the same time, because Bloomberg, Reuters and the wires write headlines from it and those headlines become the first news sources engines cite.
Days 14 to 7: what gets rehearsed?
The CEO and CFO run a full mock call, including the hostile questions on margin pressure, customer concentration, executive departures, competition and regulation. A fumbled answer in the mock is the answer an engine will quote if it is repeated on the real call. Prepared remarks should run 12 to 18 minutes across the CEO and CFO, which is EPR's working guideline rather than a published standard.
Days 7 to 1: what gets confirmed?
The IR team confirms where the transcript will live and publishes it on the company IR page as crawlable HTML, not a PDF and not a gated replay. It briefs the CEO on one fact: everything said on the call becomes source text for answers about the company. It also plans the 72-hour post-call window before the call happens.
How should the CEO run the call?
The CEO should open with the headline result, attach a number to every key message and answer each question before bridging to the thesis. Legal boilerplate stays in its required place but does not lead the remarks.
Limit the prepared remarks to three to five key messages, each with a figure, such as revenue growth, customers added, margin change and the guidance range. A sentence like "we are pleased with our progress across multiple dimensions" has nothing for an engine to quote, so replace it with a specific claim.
Name the competitive position in the CEO's own words. If the CEO does not, an engine fills the gap from analyst commentary, press coverage or Wikipedia.
For hostile questions, use three moves: acknowledge the premise, reframe inside the thesis and give the data point. An illustrative answer: "Customer concentration is a risk we monitor. Our thesis is that depth of deployment drives expansion, and no single customer exceeded 8% of revenue." The numbers in that example are invented for illustration. Naming the analyst in the reply puts a named entity in the transcript and helps the relationship.
What happens in the 72 hours after the call?
The 72 hours after the call set most of the record that engines retrieve, because coverage, analyst notes and retail commentary all appear in that window.
Hours 0 to 4: what goes live first?
The company transcript goes live in HTML within four hours. If it is not live, engines cite the Seeking Alpha or Quartr version, which can carry commentary the company did not write. The team also reads the headlines Bloomberg, Reuters, WSJ and CNBC ran and, if one is wrong, calls the reporter in the narrow correction window.
Hours 4 to 24: who needs a call?
The IR team follows up with sell-side analysts, since their notes feed summaries on Seeking Alpha and Bloomberg Intelligence. A note that echoes the company's thesis reinforces the answer, and a note that reframes it creates a competing source. The CEO posts a short LinkedIn recap that reuses the wording from the prepared remarks.
Hours 24 to 72: what gets updated?
The IR fact sheet is updated in HTML with the new revenue, margins, guidance and key metrics. Wikipedia figures such as revenue and employee count are updated with proper sourcing, since a stale entry produces a stale answer. The team also watches Seeking Alpha, Reddit and StockTwits threads and makes sure the company's wording is present through its own transcript, release and CEO post, not through direct retail engagement that could raise Regulation FD problems. See how Reddit and FinTwit feed AI summaries.
What should an IR team measure after each call?
An IR team should measure four things within 48 hours of each call.
- Ask each of the five major engines a question about the company and check whether the answer cites the new transcript.
- Compare each engine's version of the guidance against the filed wording and log any distortion or any caveat that was dropped.
- Ask a sector question such as "who are the leading companies in this sector" and record whether the company appears and in what position.
- Check whether engines name the CEO and title correctly and quote the CEO's call language instead of someone else's description of it.
Benchmark the results against peers each quarter. The list of information sources that control AI answers in investor relations shows which outlets to watch alongside the transcript.
Which earnings call mistakes cost the most?
The costliest mistakes are the ones that remove the company's own wording from the sources engines read.
- Publishing the transcript as a PDF or behind a JavaScript portal, which many AI crawlers cannot read.
- Reading the press release aloud, which wastes call time and adds no new content to the transcript.
- Answering a hostile question with "let me step back," which reads as evasion in the transcript.
- Letting the narrative frame change every quarter, which fragments how the company is described. Consistent wording across quarters, the filing and the deck is covered in AI-readable disclosures.
- Skipping the Wikipedia and fact-sheet updates after each call.
