How to search earnings call transcripts (and why keyword search fails)
Everyone starts with keyword search, because every tool offers it. You type "supply constrained", get a list of hits, and assume the ones you did not get do not exist. That assumption is where most transcript research quietly goes wrong, and it is worth understanding why before you trust a result that came back thin.
Executives do not use your words
A CFO does not say "problem". They say headwind, pressure, softness, a challenging environment. They do not say "we are running out of capacity"; they say demand is outstripping supply, or lead times have extended, or we are allocating. Search for the plain word and you find the small subset of speakers who happened to use it.
This is not a flaw in any particular tool. It is the nature of matching strings against a language where the same fact has a dozen acceptable phrasings and the professional register avoids the blunt ones. The tool did what you asked. What you asked was narrower than what you meant.
Three failures worth recognising
- The false negative: three results come back, you conclude the topic barely came up, and the other forty mentions used a different phrase.
- The lost subject: a search for "margins" across an industry returns hundreds of passages with no way to tell which company each belongs to without opening every one.
- The countless list: you find who mentioned something but not how many did, so a handful of quotes gets mistaken for a trend.
Ask for the thing, not the word
The better unit of research is a question with its filters attached: what did companies in this sector say about pricing power this quarter, which of them quantified it, and how many were there in total. That is three different operations against the archive, and typing one keyword performs none of them properly.
A question also survives being wrong. If the first phrasing finds nothing, the next attempt can use the wording an executive would actually choose, and the one after that can widen the window. Keyword search gives you a result set; a question gives you a search that can be corrected.
Counts and quotes are different jobs
A quote proves that something was said. A count proves how widespread it was. Confusing the two is how research turns into anecdote: five vivid passages about a theme feel like a trend, and they are five passages out of a population you never measured.
Any serious claim about a sector needs both. The count establishes that a pattern exists; the quotes show what the pattern actually sounds like and let a reader check. An answer with only one of the two should be treated as half an answer.
How earnings.chat helps
You ask in plain language and the retrieval is handled behind it: the archive is searched, the scope is counted, and the answer comes back with the passages that support it, each labelled with the company, the date and the speaker. When the first phrasing finds nothing, the search is retried with the words an executive would use rather than handing you an empty list.
The archive covers more than 253,000 earnings calls from 12,853 companies back to 2020, and questions can be scoped by sector, industry, country or period. What it will not do is fill a gap with something plausible: if the calls do not say it, the answer says the calls do not say it.
252,000+ earnings calls in the knowledge base, answers with verbatim quotes and sources, new calls within minutes.
Open earnings.chat