Can AI predict CPI? We replayed the night before the September 2026 report: Sikt hit headline inflation at 3.7% but missed core at 2.6%. Here is why.
By Kristian FagerlieFounder & CEO, Sikt Intelligence
Short answer: partly. We set Sikt Intelligence's clock back to midnight on October 14, 2026, hours before the September CPI report, and asked for both inflation numbers. It called headline inflation exactly at 3.7%, a tenth above the economists' consensus. It missed core inflation: it expected 2.4–2.5% and gave only about 6% to the actual 2.6%. That pattern fits what professional forecasters know well: headline CPI is easier to call when gasoline drives it, and core CPI is harder. One replayed release proves very little, and this was a look back, not a live forecast.
The consumer price index (CPI), published monthly by the Bureau of Labor Statistics (BLS), tracks the prices people pay for a fixed basket of goods and services: food, fuel, rent, cars, airfares, medical care and more. The number most people quote is the headline rate: how much the whole index rose over the past twelve months.
Core CPI leaves out food and energy, the two most volatile parts. Economists and the Federal Reserve watch it because it shows the underlying trend better. Shelter, meaning rent and the estimated rent of owned homes, is the heaviest single weight, at about 35.6% of the index as of December 2025 (BLS).
In September 2026 the two measures told very different stories (BLS):
Economists had expected 3.6% for the headline rate, and 2.5% (0.2% in the month) for core (Trading Economics, CNBC). Both came in hot, and the report briefly pushed up market odds of an October Fed hike (see will the Fed raise rates in October?).
Professional forecasters use a few standard tools:
So in a month like this one, the energy-driven part of headline inflation is close to known before the report. The surprises come from the many small prices inside core.
Sikt Past is a new Sikt feature that works like a time machine for forecasting. You pick a moment in the past, and Sikt can only read the news pages we saved before that moment. Every page is stamped when it is saved, so nothing from the future can slip in. You can then ask it questions as it was at that moment.
In the video, we set Sikt's clock to midnight on October 14, 2026, and asked: the US releases its CPI numbers tomorrow; what will headline CPI and core CPI be, year over year? Every source Sikt used was dated October 13 or earlier. Its answer, next to the consensus and the result:
| September 2026 CPI | Sikt (midnight Oct 14) | Economists' consensus | Actual | Result |
|---|---|---|---|---|
| Headline, year over year | 3.7% | 3.6% | 3.7% | Hit |
| Core, year over year | 2.4–2.5% (about 6% on 2.6%) | 2.5% | 2.6% | Miss |
On headline inflation, Sikt landed exactly on the number and beat the consensus by a tenth. On core it was in line with the economists, and wrong in the same direction: it put only about 6% on the outcome that happened. Sikt did not see the core surprise coming.
Two honest caveats. This is a retrospective replay: we ran it after the report, even though Sikt could only read what existed before it. And it is one release. A good forecaster must be scored over many releases with a proper rule such as the Brier score, not judged on one hit.
We cannot know for sure why the result split this way, but the likely reason is the one above. Headline inflation in September was driven by gasoline, and gasoline prices were in the news every day: anyone reading the coverage closely could estimate most of the 0.6% monthly rise in advance. Economists expected a monthly rise of that size too.
Core inflation has no such daily signal. It is the sum of hundreds of prices, from rent to airfares (up 22.4% over the year in September, per Trading Economics), and a 0.1-point miss in the monthly rate is enough to move the yearly rate. Core held at 0.3% in the month instead of slowing to the expected 0.2%.
The miss was not unusual, but it is still a miss: to add value on CPI, a forecaster has to find signals on core that the consensus is missing.
Where it helps:
Where it does not:
For the bigger picture, see AI forecasting, explained.
The October 2026 CPI report is due on Tuesday, November 10, 2026, at 8:30 a.m. Eastern (BLS schedule). It comes after the Fed's October 27–28 meeting, so it will shape expectations for the last meeting of the year in December, and with it the debate on recession odds. Gasoline prices will again matter most for the headline number, and the open question is whether core keeps running at 0.3% a month.
We plan to forecast that release with Sikt live, before the numbers come out, and to score it afterwards, hit or miss.
Sikt Intelligence is an AI forecaster for real-world events: rate decisions, inflation reports, elections and geopolitical flashpoints. It researches the news, gives an honest probability next to the market's price, and is scored when the answer arrives. You can follow it live on the odds pages and the 2026 midterms. Sikt is in research; join the waitlist for early access. Nothing here is financial or investment advice.
The October 2026 CPI report comes out on Tuesday, November 10, 2026, at 8:30 a.m. Eastern. The November report follows on December 10, 2026 (BLS).
Partly. AI can read gasoline prices, previews and Fed commentary quickly and give a probability for each outcome, and in our replay it called September's headline CPI exactly. But it sees the same public data as economists, and core inflation surprises are hard for everyone. Judge any AI inflation forecast over many releases, not one.
Headline CPI covers everything in the basket. Core CPI leaves out food and energy, which swing a lot from month to month, so it shows the underlying trend. In September 2026, headline inflation was 3.7% and core was 2.6%, because energy prices rose much faster than everything else.
Prices rose 0.6% in the month and 3.7% over the year, up from 3.4% in August. Core CPI rose 0.3% in the month and 2.6% over the year, up from 2.4% (BLS).
Will the Fed raise rates on October 28? Sikt's live AI forecast next to Kalshi and Polymarket odds, updated before the meeting and scored after it, with the reasons behind it.
Read the article
Live Kalshi and Polymarket odds of a US recession in 2026 and 2027, what the latest GDP and jobs data show, and how these markets define a recession.
Read the article