> ## Documentation Index
> Fetch the complete documentation index at: https://decageo.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.
# Did Your ChatGPT Ranking Drop Because of the Model? It Changed Once in 22 Weeks
> Probably not the model. DecaGEO's ChatGPT model changed once in 22 weeks. In the other weeks, about 1 in 4 ranked brands fell 3+ places every week.
**Between May 10 and September 27, 2026, if your drop did not land in the week of July 26, the model that measured it was the same in both weeks. From May 3 to September 27, 2026, DecaGEO's ChatGPT rankings ran on GPT-5.4 and then GPT-5.6 Terra, and the switch happened once, in that week, in all ten categories at the same time. In the 20 week-to-week comparisons without a switch, between 20.8% and 27.8% of the brands ranked in both weeks fell three or more places. In 979 of those 1,586 drops (61.7%), ChatGPT recommended the brand in fewer answers than the week before, and the position points behind its DECA Score fell with them. In 102 (6.4%), the brand held on both counts and was passed by brands that rose. Chart History and the answers behind it show which of these happened to yours.**
## Check which model measured each week
In Chart History, the weekly table has an **AI Platform** column that names the engine and model version behind each week's numbers. Across the 22 weeks from May 3 to September 27, 2026, it shows one boundary. Weeks up to July 19 show gpt-5.4 (GPT-5.4), and weeks from July 26 show gpt-5.6-terra (GPT-5.6 Terra). Each category kept the same question set in all 22 weeks.
If the week of your drop and the week before it show the same model, the drop happened under the same measurement conditions. Every week in September 2026 was measured on GPT-5.6 Terra, so a September drop falls in this group.
## A three-place drop happens to about a quarter of ranked brands every week
With the model and the questions held fixed, rankings still move. In every one of the 20 same-model comparisons, between 20.8% and 27.8% of the brands ranked in both weeks fell three or more places, and between 18.6% and 32.2% of the previous week's ranked brands dropped off their category ranking.
## What the model change week did
The week of July 26 moved more than any same-model week on four of five measures. On drops of ten places or more it stayed inside the usual range.
| Share of brands that | 20 same-model weeks | Week of July 26 |
| - | - | - |
| Fell 1+ place | 36.1–44.6% | 46.6% (132 of 283) |
| Fell 3+ places | 20.8–27.8% | 32.9% (93 of 283) |
| Fell 5+ places | 13.6–19.1% | 24.4% (69 of 283) |
| Fell 10+ places | 4.7–11.3% | 7.8% (22 of 283) |
| Dropped off the ranking | 18.6–32.2% | 36.4% (162 of 445) |
*DecaGEO, ChatGPT, US, English, 10 B2B software categories, May 3 to September 27, 2026. "Fell" rows count brands ranked in both weeks. "Dropped off" counts brands ranked in the previous week. The week of July 26 compares GPT-5.4 (July 19) with GPT-5.6 Terra (July 26).*
On the one-place measure, the week of July 26 had 132 drops, fewer than the highest same-model count, 145 in the week of May 17. Its share was higher because 162 brands dropped off, more than in any same-model week, so only 283 brands were ranked in both weeks, against 299 to 343 otherwise. Drops of three places or more rose in count as well as in share. The 93 in the week of July 26 exceed the highest same-model count, 92 in the week of July 5, and as a share of the smaller group they come to 32.9%. The 69 drops of five places or more likewise exceed the highest same-model count, 61 in the week of June 28.
By category, the share that fell three or more places was above that category's same-model range in 4 of the 10 categories and inside it in the other 6.
If your drop is in the week of July 26, set that week aside and compare the weeks after it with each other.
## What changed in the brand's own numbers
For the 1,586 same-model drops of three or more places, two of the brand's own numbers show why the rank fell. One is the count of the week's answers that recommended the brand. The other is its position-weighted total, the sum of the points DECA Score gives each recommendation by its place in the answer. A rank can fall because ChatGPT recommended the brand in fewer answers, because it recommended the brand as often or more but the position-weighted total fell, or because the brand held both while other brands gained.
| The brand's own numbers vs. the week before | Drops | Share |
| - | - | - |
| Fewer answers recommending the brand, lower position-weighted total | 979 | 61.7% |
| As many or more answers recommending the brand, lower position-weighted total | 482 | 30.4% |
| Fewer answers recommending the brand, same or higher position-weighted total | 23 | 1.5% |
| Both the same or higher | 102 | 6.4% |
*DecaGEO, ChatGPT, US, English, 10 B2B software categories. Brands ranked in both weeks that fell 3 or more places, in the 20 week pairs measured on one model, May 3 to September 27, 2026.*
In 1,484 of the 1,586 drops (93.6%), at least one of the brand's own numbers fell. Most of the time fewer answers recommended the brand. In another 482 cases as many answers recommended it or more, but its position-weighted total fell. Holding both counts and still being passed happened in 102 drops.
## How to read your own drop
1. **Check the AI Platform column for both weeks.** If the model differs, the drop spans a model change (from May 3 to September 27, 2026, only the week of July 26). Compare the weeks after it with each other.
2. **Check whether your [Mention Rate](/docs/metrics/mention-rate) fell.** Mention Rate is the share of the category's answers that recommend you. If it fell, fewer answers recommended you. The Prompt & AI Response tab lists the week's prompts whose answers recommended you and the brands each answer named.
3. **If Mention Rate held, check your position in the answers.** The Brand Ranking Trail in Prompt & AI Response shows every recommended brand in answer order, with yours at its position. The points you earn from an answer depend on your place in its list and on how many brands the list names, as the [DECA Score](/docs/metrics/deca-score) example shows. In this week's answers that recommend you, note your place in each list and how many brands that list names.
4. **If both held, look at who moved above you.** DECA Score is measured against the week's top brand in the category, so it can move when the leader's total moves even if yours did not. The category ranking's weekly movement shows which brands climbed past you.
5. **Look at the following week before acting on one week.** Of 1,430 same-model drops of three or more places whose following week was measured on the same model, 265 (18.5%) returned to within one place of where they started. Of all 1,498 same-model drops with a following week, including those whose following week was the model change week, 871 (58.1%) climbed back at least one place.
DecaGEO's public category rankings show every brand's rank, weekly movement and Mention Rate without signup. The free plan adds Chart History for the last four weeks with the AI Platform column, and lets you track one product. Pro keeps the full chart history, which reaches back across the July 26 boundary, and adds weekly change alerts for unlimited tracked products.
## FAQ
### Did ChatGPT's model change cause my AI ranking to drop?
Between May 3 and September 27, 2026, only a drop in the week of July 26 can be tied to a model change. In that window, the model behind DecaGEO's ChatGPT rankings changed once, from GPT-5.4 to GPT-5.6 Terra in the week of July 26. Chart History's AI Platform column shows the model for every week. See [Check which model measured each week](#check-which-model-measured-each-week).
### How often do brands drop in AI rankings when nothing changes in the measurement?
Every week. In the 20 week pairs measured on one model, 20.8% to 27.8% of brands ranked in both weeks fell three or more places. See [the weekly chart](#a-three-place-drop-happens-to-about-a-quarter-of-ranked-brands-every-week).
### ChatGPT still recommends my brand as often, but my rank dropped. Why?
Either your position-weighted points fell in the answers that recommended you, or you held both and other brands gained. Among same-model drops of three or more places, 482 had as many answers recommending the brand or more but a lower position-weighted total, and 102 held both. See [What changed in the brand's own numbers](#what-changed-in-the-brands-own-numbers).
### Should I react to a one-week drop in my ChatGPT ranking?
Check the next week first. Of 1,430 same-model drops of three or more places whose next week was on the same model, 18.5% returned to within one place of their earlier rank. Of all 1,498 with a following week, 58.1% climbed back at least one place. See [How to read your own drop](#how-to-read-your-own-drop).
## How we measured
The panel's questions, schedule and metric definitions are on the [methodology page](https://decageo.ai/methodology). What is specific to these figures:
* **Window.** 22 weekly rankings, weeks of May 3 to September 27, 2026, ten categories, 21 pairs of consecutive weeks. The week of July 26 is the only pair measured on two models (GPT-5.4, then GPT-5.6 Terra). The other 20 pairs are the same-model weeks.
* **Fell N places.** A brand ranked in the same category in both weeks whose rank number rose by N or more. Brands with equal DECA Score share a rank and the next places are skipped. Shares divide by the number of brands ranked in both weeks.
* **Dropped off.** A brand ranked in a category one week and absent from that category's ranking the next. The share divides by the number of brands ranked in the first week.
* **Own numbers.** Answers recommending the brand is the count of the week's answers that recommended it as a solution. As on the [Mention Rate](/docs/metrics/mention-rate) page, name-drops, negative references and comparison filler are not counted. Position-weighted total is the sum of the position points behind its DECA Score. "Same or higher" includes equal values.
* **Following week.** For drops in the weeks of May 10 to September 20, the brand's rank one week after the drop. "Returned" means within one place of the starting rank. A brand absent from the ranking that week counts as not returned. For the 68 drops in the week of July 19, the following week is the model change week. The 1,498 include them and the 1,430 do not.
Related: [Reading Chart History](/docs/using/chart-history) · [Rank & Weekly Movement](/docs/metrics/rank-and-weekly-movement)