Published September 29, 2026 · Last updated September 29, 2026 · Based on a September 25, 2026 study — this page is re-checked monthly.

ChatGPT Recommended My Business Last Week — Why Am I Gone This Week?

Short answer: Because ChatGPT doesn't keep a list — it rebuilds the answer every time. A Spearleaf study (published September 25, 2026) asked the identical question again 4–9 days later in the same markets and found only 48.2% of the businesses shown the first time were shown again (469 of 974). The top spot held even less often: the same business ranked first both times in just 34.1% of markets (73 of 214). You didn't break anything. The recommendation didn't "stick" because recommendations aren't stored — they're regenerated, and asking the same question again is enough to produce a different set of names.

The study that measured the churn

Spearleaf ran a repeat-capture study across 216 U.S. markets — 12 service categories across 18 cities, captures from August 2026, "one plain question in every market, the kind a real buyer types." In 214 paired markets, they asked the identical question a second time, 4 to 9 days later, and compared the names.

The result is the sentence every contractor needs taped to the wall: "Nothing announced the turnover, and no interface change accompanied it. Asking the same question again was enough to produce a different set of names."

Of the 974 businesses ChatGPT displayed the first time, only 469 — 48.2% — were displayed again. The #1 recommendation was even less durable: the same business held first position both times in only 73 of the 214 markets — 34.1%. Read that as a contractor: if you were the named shop on Tuesday, the odds say you're not the named shop next Tuesday, and it has nothing to do with your work.

You're not being punished — there's no list to fall off

The natural owner reaction is to hunt for what changed: did my reviews drop, did my Google profile get flagged, did I get outbid? But the study's own finding rules out that whole frame. No announcement, no interface change, no signal from the business side — the question alone, asked again, shuffled the deck.

Think of it this way. Google's local pack is a dispatcher with a roster: it keeps a standing list and adjusts it. ChatGPT's recommendation panel is a dispatcher who picks a fresh crew on every call from whoever looks available right now. You didn't get fired from a roster you were never on. You were picked for one job and not the next.

Google keeps a list; ChatGPT rebuilds one

The study included the comparison that makes this click. Wave-to-wave overlap — how much the same names appear from one capture to the next — was 0.82 for Google's local pack, 0.36 for the ChatGPT panel, and 0.26 for AI Overviews.

Sit with that. Google's pack is more than twice as sticky as ChatGPT's panel (0.82 vs. 0.36). Google maintains something like a persistent ordering of local businesses; ChatGPT drafts a fresh panel each time. For a restoration contractor, the practical translation: your Google Business Profile and your local rankings are the durable floor. The ChatGPT panel is weather. Don't confuse the two, and don't spend floor money chasing weather. (See how the different AI engines choose contractors for why the engines behave so differently.)

The top spot is the least stable seat in the room

The 34.1% number deserves its own paragraph, because it's the one that will cost owners the most money if they misunderstand it. Being named somewhere in the panel reoccurred for roughly half the businesses. Being named first reoccurred for roughly a third.

That's the pattern that feeds the snake-oil economy: a vendor shows you a screenshot where your shop is #1, you sign the retainer, and by the second week you're gone — not because the vendor failed, but because the #1 slot was a coin flip all along. Nobody can sell you the top of a panel that regenerates itself.

It's not one study's quirk — the corroboration agrees

An independent study from Evoke Strategy in Tampa, Florida (September 18, 2026) measured the same instability at a different scale: 28,488 Florida businesses across 9 cities, 18 industries, and 4 AI platforms. Only 1,676 — 5.9% — ever showed up at all. And of those that did, 43.2% surfaced in only one of 12 searches, 22.4% in two, and just 12.9% in five or more.

Two different studies, same verdict: showing up once is common; showing up repeatedly is rare. As Evoke Strategy's Devon Vocke put it: "The bigger question is whether your company is repeatedly discoverable across the different questions customers are asking and the different AI platforms they're using."

What "stickiness" actually means for a contractor

• Stop thinking about a position; start thinking about a batting average. You cannot lock a spot in an answer that gets rebuilt from scratch. What you can do is raise your probability of being in the drawn pool on any given ask.
• Measure the pattern, not the moment. One screenshot is one roll of the dice. Four weekly runs of the same prompt is a batting average — and it's the average you should be managing.
• Treat disappearance as data, not defeat. If you were named in week one and gone in week two, you're experiencing the 51.8% majority outcome, not a failure. The businesses to worry about are the ones that never appear at all — in the Evoke data, that's 94.1% of them. (If you're not appearing at all, start with why your business doesn't show up.)

What to actually do this week

1. Stop trusting a single screenshot — yours or a vendor's. One capture tells you almost nothing about next week. Don't make decisions — and definitely don't sign contracts — off one.
2. Start a weekly churn log. Same prompt, same city, same wording, once a week. Log which businesses get named and who holds the top spot. Four weeks of that is your real visibility picture. (Fold this into your monthly AI search visibility check.)
3. Interrogate any "we'll get you into ChatGPT" pitch with the churn question. Ask the vendor: "Show me the named set across four weekly runs, not one screenshot." A vendor who can't — or who promises a locked spot — is selling the 34.1% coin flip as a guarantee.
4. Log the competitors' churn too. The rival who replaced you this week will very likely vanish next week — the same 48.2% math applies to them. (If the same rival keeps showing up week after week, that's a different problem — see when ChatGPT recommends your competitor.)
5. Keep building the durable floor. Reviews, accurate listings, pages that answer buyer questions in quotable sentences — these raise your odds of being drawn in every rebuild. They don't lock the spot (nothing does), but they weight the dice.

The line to remember

Being named once by ChatGPT is a weather report, not a ranking. The panel rebuilds itself every time — 48.2% reappear, the top spot holds 34.1% of the time — so watch the pattern across weeks and act on the pattern, never on a single answer.