Tracking AI Mentions in Earnings Calls: A Practical Guide for Staying Ahead of the Market

When a technology shift is underway, executives talk about it on earnings calls long before it shows up cleanly in the financial statements. Artificial intelligence is the clearest current example: capex plans, product roadmaps, and hiring priorities all surface first as spoken commentary. Anyone who systematically monitors how often - and in what context - companies discuss AI has a measurable head start on investors who wait for the numbers.

The good news: with transcripts available as structured data, building that monitoring pipeline is a weekend project, not a research department.

Why AI Mentions Are a Signal Worth Measuring

An earnings call is one of the few recurring events where management explains strategy in its own words and then defends it under analyst questioning. When AI comes up in that setting, it usually points at something concrete: a new spending commitment, a product bet, a competitive worry.

Measured across many companies and quarters, those mentions become useful in several ways:

Early innovation signal. A jump in AI references frequently precedes announcements about capex, partnerships, or new product lines.
Confidence barometer. Whether executives sound enthusiastic, defensive, or evasive about AI tells you as much as the mention count itself.
Relative positioning. Comparing mention trends across an industry shows which players are moving first and which are reacting.
Downside detection. Companies that barely acknowledge AI in a sector where every peer discusses it may be falling behind - a risk flag in its own right.

Left unmeasured, all of this stays anecdotal. Turned into data, it becomes something you can chart, backtest, and alert on.

Defining What Counts as an "AI Mention"

Before writing any code, decide what your keyword universe looks like. "AI" alone is too narrow and too noisy at the same time. A workable starting set:

artificial intelligence / AI
machine learning
deep learning / neural networks
large language models / generative AI
automation
predictive analytics

The raw hit count is only step one. A mention inside a cautious risk-factor answer means something different from one inside a product announcement, so you'll want to keep the surrounding context - ideally the full speaker segment - attached to every match.

Building the Tracking Workflow

Step 1: Get transcripts as structured data

Scraping IR pages or transcribing webcasts by hand doesn't scale past a handful of tickers. An API that serves transcripts as JSON - with speaker attribution - is the foundation. EarningsAPI exposes 250,945 earnings calls across 12,728 companies, broken into 11.9 million individual speaker segments, so a mention can be attributed to a specific CEO answer rather than just "somewhere in the call." Endpoint details live in the API docs.

Step 2: Run keyword queries across the corpus

With full-text search over the transcript archive, counting AI references per company per quarter is a single query loop. Store the results in a simple table: ticker, quarter, term, hit count, and the matched segments.

Step 3: Read the context, not just the count

A spike in mentions is an invitation to look closer, not a conclusion. Feed the matched segments through a sentiment or classification step - even a lightweight LLM prompt works well here - to separate "we are shipping AI features" from "AI poses risks to our licensing model."

Step 4: Watch the trend line

Quarter-over-quarter change is where the signal lives. Is a retailer suddenly discussing machine learning three times as often? Is an entire sector's mention curve bending upward? Trend inflections tend to lead press coverage by one or two quarters.

Step 5: Cross-check against fundamentals

Correlate mention trends with R&D spend, capex, and revenue growth. Companies whose AI talk is backed by rising investment deserve different treatment than those where the language outruns the budget.

Tips That Make the Pipeline More Useful

Automate the refresh. Poll for new transcripts after each reporting day so your dataset never lags the market.
Use an LLM as your analyst. If you work in Claude or another MCP-capable assistant, you can query the transcript corpus conversationally through the EarningsAPI MCP server at mcp.earningsapi.io - "which industrial companies increased AI mentions most this quarter?" becomes a one-line question. Setup is covered in the MCP guide.
Don't only watch mega-caps. Mid-caps and niche players adopting AI early are where mention data adds the most information, because nobody else is covering them.
Refresh your vocabulary. The terminology moves fast - "agentic," "inference costs," and "copilot" barely existed in transcripts three years ago. Revisit your keyword list every few quarters.

Mistakes to Avoid

Treating frequency as quality. Some management teams sprinkle AI into every answer as theater. Weight mentions by specificity - named products, dollar figures, and timelines beat vague enthusiasm.
Dropping the tone. A high mention count driven by risk disclosures is a bearish datapoint dressed as a bullish one.
Applying one lens to every sector. AI in a hospital operator, a bank, and a machine-tool maker are three different stories. Calibrate expectations per industry.
Stopping at the keyword hit. Always read (or have your model read) the full segment before acting on a match.

Wrapping Up

AI commentary in earnings calls is a leading indicator hiding in plain text. Companies telegraph their technology strategy every quarter - the only question is whether you're set up to capture it systematically. With transcripts available as clean, speaker-segmented JSON, a mention-tracking pipeline is one of the highest-leverage small projects a data-driven investor or developer can build.

If you want to start today, EarningsAPI provides the transcript corpus, full-text search, and MCP access to build on - check whether your tickers are covered with the coverage check on the homepage.