S&P 500 Companies Citing AI on Earnings Calls: Flat Talk, Moving Profit

FactSet found the term "AI" on 331 of the 493 S&P 500 earnings calls held in the Q2 2026 reporting window. That is 67% of the index, and it is no higher than the quarter before. Over the same stretch, reported S&P 500 earnings growth swung from 13% to 50% and back to about 30%. The share of S&P 500 companies citing AI on earnings calls stood still while profit moved, which tells you the mention itself has stopped carrying information.

What the FactSet AI count actually measures

The number is simpler than the coverage suggests. Each quarter, FactSet senior earnings analyst John Butters searches the conference call transcripts of every S&P 500 company that held an earnings call during a set window. A call counts if the term "AI" appears in it. One passing mention counts the same as a 40-minute product briefing.

The windows are not calendar quarters. The Q2 2026 count covered calls from 15 June to 14 September 2026. The Q1 2026 count covered 15 March to 11 June. Each window catches the reporting season for that quarter, plus a few off-cycle reporters whose fiscal years end in other months.

So the figure is a measure of vocabulary, not of revenue, spending or profit. It answers one question well: how many large US companies feel they need to say the word to their investors. That is a useful cultural indicator. It is a weak financial one, and the rest of this post shows why.

FactSet reports the count against the number of calls actually held, not against 500. The Q2 2026 total was 493 calls, Q1 2026 was 498 and Q4 2025 was 485. A raw count can rise simply because more companies reported inside the window. The share is the cleaner series, and the share is where the plateau shows most clearly.

Five quarters of S&P 500 AI mentions, in one ledger

The table below lines up every count FactSet has published since mid 2025, next to FactSet's own blended earnings growth rate for the index in the same quarter. The growth rates are snapshots taken during each season, and the column states how much of the index had reported at the time.

Counts are as first published in each FactSet note. Two were restated upward in the following quarter's note without explanation. Growth rates are blended, mixing reported actuals with estimates for companies still to report, so none is final. Last verified 26 October 2026.

The AI mention count has stopped rising

From Q2 2025 to Q4 2025 the count added 39 calls in two quarters. From Q4 2025 to Q2 2026 it added none. FactSet's own framing in September was that Q2 2026 marked the third straight quarter above 65%, which is accurate and also a polite way of describing a ceiling.

The sector detail explains where the ceiling sits. Information Technology hit 97% of calls in Q2 2026, with Financials at 91% and Communication Services at 90%. Those sectors cannot climb much further. Information Technology and Financials alone supplied 139 of the 331 citing calls, or 42%, by our calculation from FactSet's sector counts.

The gap to the averages is a lagging artefact

Headlines often lead with the multiple over the historical average. The Q2 2026 count is 1.9 times the 5-year average of 178 and 2.9 times the 10-year average of 114. Those ratios are shrinking, not growing, because the averages are catching up with the recent surge. The 5-year average rose from 136 in the Q3 2025 note to 178 in the Q2 2026 note.

That matters for how you read the next headline. A "record versus the average" story can keep running for years after the underlying count has gone flat. My view is that the average comparison has become the least informative line in each FactSet note.

About 162 S&P 500 companies held a Q2 2026 call without saying "AI". It is tempting to read them as behind. Some are, but many sit in sectors where the word has no obvious product meaning yet, and saying it would invite questions they cannot answer. Silence on a call is a disclosure choice, not a strategy audit.

S&P 500 earnings growth moved while AI mentions did not

Here is the test the topic sets: does more talk show up as more profit? At the index level, the answer from FactSet's own numbers is no visible link. The mention share held at 67% to 68% for three quarters. Over the same three quarters the blended growth rate moved from 13.2% for Q4 2025 to 27.1% for Q1 2026 and 50.4% for Q2 2026.

A flat input next to a quadrupling output is not evidence that the input caused anything. It is the opposite pattern from the one the "AI is driving earnings" story implies. If mention counts tracked profit, the count would have had to jump alongside growth, and it did not.

What drove Q1 2026 instead

FactSet was specific. In its 1 May 2026 update the Q1 blended rate jumped from 15.0% to 27.1% in a single week. The note named positive earnings surprises from three Magnificent 7 companies, Alphabet, Amazon and Meta Platforms, as the largest contributors. Three companies moved the index rate by 12 points in a week. A count of 337 calls had nothing to do with it.

With 24% of the index reported, FactSet's 23 October 2026 update put blended Q3 growth at 29.5%, with revenue growth of 12.7%. That would be the third straight quarter above 25%. It is still a strong number, and it is a long way below Q2's 50.4%.

Where the Q2 2026 AI profit actually came from

The Q2 quarter is the most useful single case, because FactSet decomposed it. The 50.4% blended rate would be the highest since Q2 2021. Two companies explain most of the excess.

According to FactSet's 7 August 2026 update, Alphabet's GAAP earnings included a $98 billion gain in other income, primarily net unrealized gains on equity securities. Amazon's included a $53.4 billion gain in other income, primarily due to its investments in Anthropic. Excluding Alphabet and Amazon, the index growth rate falls to 32.0%.

The original figure in this post

The working is short. 50.4% minus 32.0% is 18.4 percentage points. Divided by 50.4%, that is 36.5% of the quarter's reported growth rate attributable to two companies. Add the two gains together and you get $151.4 billion of other income, booked as marks on equity stakes rather than as sales of AI products.

Note the precise claim. Excluding Alphabet and Amazon removes their whole earnings change, not only the gains, so 18.4 points is an upper bound on the gain effect. FactSet's own wording ties the gains to equity securities and Anthropic. It does not publish a gains-only adjustment.

Why this matters for the "more talk, more profit" story

The single largest AI effect on S&P 500 earnings in Q2 2026 was not a customer buying an AI product. It was two investors marking up stakes in private AI labs. That is a profit from AI valuations, and it reverses if those valuations fall. The way private lab valuations are being set is covered in the comparison of Anthropic and OpenAI valuation multiples.

I would treat any index growth figure that includes those marks as a statement about private market prices, not about AI adoption. Quarters like Q2 make the mention count look connected to profit, when the connection runs through a revaluation instead.

AI citers vs non-citers: FactSet measures price, not profit

The comparison most readers assume exists does not. FactSet does not publish earnings growth for companies that cite AI against those that do not. It publishes average share price changes. That is a reasonable thing to track, and it answers a different question: whether investors reward the word, not whether the word earns anything.

Each row is measured from 31 December of the year in progress, so the Q4 2025 row covers only ten weeks. Spreads are Zan Digital calculations from FactSet's published averages. Medians appear only where FactSet stated them.

Averages flatter the citers

The Q1 2026 note is the one to read closely. On average, citers gained 13.1% against 7.7% from 31 December 2025. On a median basis, citers gained 5.5% against 6.2% for non-citers. A higher average with a lower median means a few very large winners pulled the citer average up. The typical citing company did slightly worse than the typical silent one.

That is the strongest single data point against reading the AI mention as a signal. The premium belongs to a handful of companies, most of which would be on the citer list whatever they said on their calls.

Where this argument is weakest

This post argues the mention count no longer predicts profit. Here is where that case is thin, stated with the same care.

Nobody has run the earnings comparison, including us

The cleanest test would split the 493 companies into citers and non-citers and compare their own earnings growth. FactSet has not published that split, and we have no first-party earnings dataset to build it. Everything above compares a flat index-level count with index-level growth. That shows the count does not track the aggregate. It cannot rule out a real earnings gap between the two groups.

The composition problem cuts both ways

Information Technology cites AI on 97% of calls and was also among the fastest-growing sectors in Q2 2026, per FactSet's sector ranking. Without the Alphabet and Amazon gains, the other 498 companies still grew earnings by 32.0%. Some of that is real AI revenue, mostly for sellers of chips, servers and cloud capacity. For those companies, the talk and the profit genuinely line up.

So the fairer version of the claim is narrower. The AI mention distinguishes sectors, and sector explains a lot of earnings growth. Inside a sector, there is no published evidence that saying the word more earns more.

The counts themselves move

FactSet restated Q3 2025 from 306 to 314 and Q4 2025 from 331 to 334 in later notes, without explaining why. Late transcripts and index membership changes are the likely causes, but that is an inference. A three-call revision does not change the conclusion. It does mean no single quarter's count should be quoted to the unit.

What to watch when the Q3 2026 AI count lands

The Q3 2026 count is not out yet. Based on FactSet's pattern, the window runs from mid September to early December, and the note lands soon after. The Q3 2025 note, published 8 December 2025, recorded 306 calls. That is the year-on-year baseline.

Three readings would change my view. First, a share well above 70%, which would mean the long tail of non-tech sectors is adopting the vocabulary. Second, a FactSet earnings split between citers and non-citers, which would finally test the claim directly. Third, a Q3 growth rate that holds near 30% once most of the index has reported, without a repeat of Q2's investment gains.

If none of those appear, the right reading is that the count measures how fashionable a word is. That is worth knowing. It is not worth trading on. The SEC has already started acting on AI claims that outrun the business, as logged in the record of AI-washing enforcement actions.

What this means for your own reporting

If you run a company that reports to investors or a board, the lesson is direct. Saying "AI" puts you in the two thirds, which is no longer a distinction. The disclosures that do carry information are line items with a number attached, and the ones worth publishing are ranked in the guide to which AI revenue metrics mean something. The gap between AI spending and operating profit at the company level is measured in the analysis of AI spend against EBIT impact.

Banks are the clearest example of a high-citing sector with an unclear payoff. Financials cited AI on 91% of Q2 calls, and how that spending shows up in reported results is traced in the review of bank AI spending in Q3 earnings.

Frequently asked questions

Where to start this week

Pull the transcript of your largest vendor's or competitor's latest earnings call and search it for "AI" yourself. Then search the same filing for a number attached to it: AI revenue, AI customers, AI gross margin. If the word appears a dozen times and the number never does, you have learned what the FactSet count cannot tell you.

Then put a reminder in early December for the Q3 2026 FactSet note. Read the share, not the count, and check whether FactSet adds an earnings comparison. If it does, it will be the first real test of whether the talk pays.

Related reading