I spend my working days inside a database of 253,129 earnings call transcripts — roughly 10,000 new calls a quarter, 11.99 million speaker segments, coverage back to 2020. When a phrase starts moving through that corpus, I usually see it before it becomes a headline. Right now, nothing is moving like "tariff refund."
As of August 24, with five weeks still left in Q3 2026, 476 companies have used the phrase on 490 earnings calls this quarter. That already makes it the biggest quarter on record for the term. For scale: all four quarters of 2024 combined produced 101 matches.
The cause is not a mystery. The U.S. Supreme Court ruled that IEEPA — the International Emergency Economic Powers Act — did not confer authority on the President to impose the tariffs built on it. With that, two years of tariff cost management flipped into a refund wave, and by mid-2026 the wave was flowing straight through corporate P&Ls.
The ruling, in one sentence
I don't need to summarize the legal outcome myself, because a company that lives in the middle of global trade already did it on an earnings call. On June 17, Expeditors International told listeners: "the bottom line here is we have a Supreme Court ruling in place that found that IEEPA didn't confer authority to the President to impose tariffs."
One sentence, and two years of sunk cost got reclassified into money coming back. When money comes back, executives talk about it. And when executives talk about something, it lands in my search index the moment the transcript does. What follows is what the corpus actually shows.
Eleven quarters of one phrase
Each figure below is the number of companies whose earnings calls match a full-text search for the exact phrase "tariff refund." One methodology note, and it covers every number in this post: this is keyword matching, so the counts are an upper bound that includes some incidental co-occurrences — a passing mention in a question counts the same as a ten-minute discussion. That noise floor is roughly constant across quarters, though. The trend is real.
| Quarter | Companies matching "tariff refund" |
|---|---|
| 2024 Q1 | 19 |
| 2024 Q2 | 20 |
| 2024 Q3 | 24 |
| 2024 Q4 | 38 |
| 2025 Q1 | 87 |
| 2025 Q2 | 122 |
| 2025 Q3 | 117 |
| 2025 Q4 | 104 |
| 2026 Q1 | 206 |
| 2026 Q2 | 381 |
| 2026 Q3 (through Aug 24) | 476 |
I read this as three regimes.
2024 was background noise. Nineteen to 38 companies a quarter, in a corpus that ingests about 10,000 calls per quarter. A phrase at that level barely registers above the incidental-match floor.
2025 was the cost regime. The count jumped to 87 in Q1 and peaked at 122 in Q2, then drifted down through 117 and 104 as the year wore on. Companies were talking about tariffs constantly that year — but refunds were a corner of the mitigation conversation, not the headline. The drift downward reads, to me, like tariffs settling in as a known cost rather than a live fight.
2026 is the refund regime. 206 in Q1, 381 in Q2 — 3.1x the year-ago quarter — and 476 in a Q3 that is only about 60% complete. Q3 is already running above 4x its 2025 counterpart. I don't have a clean single explanation for Q1's doubling from the transcripts alone, so I won't invent one; what's unambiguous is what happened once the ruling was in place: the two biggest quarters in the series, back to back, with the second one still open.
And no, the corpus didn't grow underneath the phrase. Volume runs at roughly 10,000 calls a quarter — Q2 2026 was 10,053 — so the denominator is flat. The share of companies saying "tariff refund" is what exploded, not the number of calls we transcribe.
Same words, opposite direction
The counts undersell the shift, because the phrase flipped polarity between 2025 and 2026.
Read the 2024 and 2025 matches and "tariff refund" lives inside cost language: companies describing what tariffs were costing them, how they were mitigating, and what they hoped to claw back. Read the 2026 matches and the same two words sit inside capital-allocation language: refunds received, refunds expected, and what management intends to do with the money.
That distinction matters if you use transcript data quantitatively. Term frequency tells you a theme is accelerating; the segments around the term tell you which direction it points. A count series that looks like smooth growth from 122 to 381 is actually two different conversations wearing the same phrase. It's the same reason quants pull text signals from the language around the numbers, not just the numbers themselves.
The question analysts ask now
Through 2025, the standard analyst question was some version of "how much will tariffs cost you, and what are you doing about it?" This summer it became "how big is the refund, and what will you do with it?" Across the Q3 calls I've read, three answers dominate.
Put it into price. On its August 20 call, Walmart reported adjusted operating income up 17.4% in constant currency and was explicit about both the source and the plan: "This includes the benefit from the receipt of tariff refunds in the quarter. And as we suggested on the last call, our intent was to deploy much of that back into price and that's what we're doing."
Reinvest it. e.l.f. Beauty on August 5: "In Q1, we received approximately $50 million of IEEPA tariff refunds. We plan to fully reinvest these funds in our business this year, largely through a combination of the pricing actions I spoke about and increased marketing investment." Fifty million dollars, routed into pricing and marketing rather than dropped to the bottom line.
Raise the guide. Target's quarter, per public reporting, carried a boost of roughly $752 million from tariff refunds, and the company raised full-year guidance on the back of it. On the cost side of the same trade, Caterpillar said its full-year tariff cost outlook should come in at the lower end of its previously expected range.
Three uses of the same windfall: price investment, reinvestment, guidance. If you hold consumer or industrial names, the odds rise every week that one of them faces this exact question on its next call.
One-time money looks like growth
Here is the part I'd flag before anyone builds a screen off this data: tariff refunds are, by construction, one-time. They land in operating income and make the quarter look better than the underlying business.
Walmart handled the disclosure the way you'd want every company to. From the same call: "our underlying profit growth was where we thought it would be, excluding this benefit."
That sentence is the tell. Companies that separate the refund from underlying performance are doing your diligence for you — they're telling you what the run rate looks like without the windfall. Companies that don't make that separation are the ones whose 2027 comparisons will lap quarters inflated by money that arrived exactly once.
The practical search, then, isn't just "who mentioned tariff refunds." It's "who mentioned them and quantified the exclusion" — and, on the other side of the ledger, "who mentioned them and didn't." Both of those are text queries, and text queries are what this corpus is for.
Where the refunds are landing
Slicing the Q2 2026 matches by sector, the top of the distribution looks like this:
- Consumer Discretionary — 210 companies
- Industrials — 173
- Health Care — 78
- Consumer Staples — 56
- Information Technology — 54
- Materials — 36
The top two are no surprise: importers of physical goods paid the tariffs, so importers of physical goods collect the refunds. The depth further down the list is what I'd point at. Health Care at 78 and Information Technology at 54 means this is not a retail story with a few industrial cameos — it's spread across most of the 11 sectors we classify.
The headlines cover Target and Walmart. Nobody is writing a story about the 300th company to receive a refund, and that long tail is exactly where a systematic screen earns its keep.
Tracking it yourself
Everything above comes from one full-text search endpoint. Here's the exact call for refund mentions since Q3 began:
curl 'https://earningscalls.dev/api/v1/search?q="tariff refund"&type=transcripts&date_from=2026-07-01' -H "X-API-Key: $KEY"
The q parameter supports "exact phrase" matching, AND/OR operators, and -negation, which is enough to build the refined versions of this screen:
"tariff refund" AND guidance— companies connecting the refund to their outlook"tariff refund" AND underlying— Walmart-style separations of one-time benefit from run rate"tariff refund" -IEEPA— refund discussion outside the ruling context
Run it weekly, diff the resulting ticker list against your holdings, and you'll know who got the question before the coverage catches up. And because keyword counts are an upper bound, do what I do: skim the matched segments before acting on a name. The excerpt is right there in the response — reading it takes seconds and kills the false positives.
If you want the full scaffolding — scheduled runs, per-ticker history, alerting — I walked through the whole pattern in build an earnings call theme tracker in Python. The complete query syntax is in the docs, and if you'd rather ask in plain English from inside Claude, the same corpus is exposed through our MCP server.
I've run this playbook before — last year I watched agentic AI spread through 2,400 earnings calls. That was gradual adoption: a theme compounding quarter over quarter as companies talked themselves into it. This one is different. A single court ruling produced the sharpest phrase inflection I've seen in this database — 104 companies to 476 in three quarters, with five weeks of the third still to go.
The refund wave is still building, and the transcripts are where it shows up first.
Search all 253,129 transcripts yourself — get an API key at earningscalls.dev.