How much AI talk, relative to how much R&D spend?

Every S&P 500 company now talks about AI in its annual filing. This tool weighs the talk against the spending. Pick a company and see the AI mention count in its two most recent 10-K filings, the AI-wash score for each filing, and which genuine AI signals the text actually contains.

Read this first. The data was compiled on October 7, 2026 from SEC EDGAR and covers each company's two most recent 10-K filings. Years shown are the years the filings were filed, which is how the source data records them. The AI-wash score is a heuristic: AI mentions per $1B of effective R&D (reported R&D, or 1% of revenue where that is larger), scaled so 200 mentions per $1B scores 100. It is not a rating of the company, and it is not AI revenue: this dataset contains no AI revenue figures, and no score here says anything about how much money a company makes from AI. Where a company has no R&D or revenue base recorded, its filing is shown as not scored, never as zero.

You get 2 free company lookups a day. The NVIDIA example below is already loaded and does not count.

Nvidia (NVDA)

Genuine AI signals in the latest filing

How the AI-wash score works

For each filing, the dataset counts mentions of AI terms (artificial intelligence, generative AI, LLM, machine learning, deep learning, NLP, computer vision, neural network, and standalone AI) across the full 10-K text. The score then compares that count to the scale of the company's research spending:

Effective R&D is the larger of reported R&D expense and 1% of revenue. The revenue floor exists because banks and many financial companies rarely report an R&D line, and they are scored on the revenue base instead.
Mentions per $1B is the AI mention count divided by effective R&D in billions.
AI-wash score is mentions per $1B divided by 2, rounded, and capped at 100. So 200 mentions per $1B of effective R&D scores 100, and 10 mentions per $1B scores 5.
Not scored: 17 of the 988 filings have no R&D figure and no revenue base recorded, so no denominator exists. Those filings show as not scored. That is a data gap, not a zero.

Score bands and the hype gap verdict

AI-wash score	Band	What it means
0 to 10	Low	AI talk is small relative to R&D-scale spend.
11 to 30	Moderate	Noticeable AI talk relative to the spending base.
31 to 60	High	A lot of AI talk for the size of the spending base.
61 to 100	Very high	AI talk far outweighs the recorded spending base.
Not scored	No base recorded	No R&D or revenue base in the data, so no gap can be computed.

The verdict line on each result is built from three things only: the latest filing's score band, the genuine AI signals detected in that filing, and the change in AI mentions versus the prior filing. A company can score high on talk and still carry real substance signals; the verdict says both when both are true.

The genuine AI signals

AI in the Business section: AI terms appear in the first part of the filing, where the company describes what it actually does, rather than only in risk factors or boilerplate.
Generative AI / LLM mentioned: the filing names generative AI or large language models specifically.
AI product language: the filing uses product wording such as AI-powered, AI-driven or AI-enabled.
Machine learning / deep learning mentioned: the filing names the underlying techniques, not just the AI label.

The signals are machine-detected from the filing text. They show the talk is attached to described products and business lines; they are not proof of AI revenue, and this dataset contains no AI revenue figures for any company.

What this is not

Nothing on this page is investment advice. A high score does not mean a company is dishonest: small companies with small R&D bases score high on modest mention counts, and some genuine AI builders report little R&D in the form this dataset reads. A low score does not mean a company does no AI work. Read the score as a prompt to open the filing, which is why every result links to the filing itself on SEC EDGAR.