> ## 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.
# Do ChatGPT Brand Recommendations Fade as Fast as AI Citations? 78% Stay a Week Later
> Not in ChatGPT's B2B software answers. In DecaGEO's weekly panel, 44% of cited URLs and 78% of recommended brands showed up again the next week.
**Not at the same speed, at least in ChatGPT's answers about B2B software. In DecaGEO's weekly ChatGPT panel, 44.0% of the URLs cited in a category one week were cited in that category again the next week. Of the brands ChatGPT recommended in a category one week, 78.3% were recommended there again the next. Three weeks out the split barely moved, 41.3% of URLs against 77.7% of brands, and 89.0% of top-10 brands across the ten categories were still in their category's top 10 four weeks later. All of these come from the weeks of July 26 to October 4, 2026, measured on GPT-5.6 Terra, and the one-week figures cover the ten week pairs in that window. The pages that back up an answer change quickly, and the brands the answer recommends change much more slowly, so a URL that drops out of the citations is not by itself a lost recommendation.**
## What the 11-day figure measures
On September 30, 2026, Profound published a study titled ["The half-life of an AI citation is 11 days."](https://www.tryprofound.com/blog/the-half-life-of-an-ai-citation-is-11-days) It found that "the median page loses half its citation share within 11 days of peaking," tracking pages daily across seven answer engines from September 2025 to September 2026. Its unit is one page on one engine, and its outcome is that page's share of the citations the engine gave in the page's category that day. The study's advice for marketers is to publish "with enough specificity to be chosen and enough consistency to be chosen again."
A page's citation share and a brand's recommendation are two different counts. The brand behind a page can keep being recommended while the pages cited next to its name change.
## Citations turn over every week, and most recommendations carry over
For each of ten B2B software categories, DecaGEO compares one week's cited URLs and recommended brands with the following week's, under the same questions and, except in the model change week, the same model.
| Week pairs | Cited URLs cited again | Recommended brands recommended again |
| - | - | - |
| One week apart, July 26 to October 4 (GPT-5.6 Terra) | 44.0% (10,292 of 23,373) | 78.3% (3,237 of 4,136) |
| Three weeks apart, same window | 41.3% (7,731 of 18,712) | 77.7% (2,578 of 3,318) |
| One week apart, May 3 to July 19 (GPT-5.4) | 39.0% (12,225 of 31,310) | 72.4% (3,541 of 4,893) |
| Model change week, July 19 to July 26 | 30.4% (900 of 2,963) | 63.6% (283 of 445) |
*DecaGEO, ChatGPT, US, English, 10 B2B software categories. Counts are summed over every category and week pair in the row. Definitions in [How we measured](#how-we-measured).*
In all 22 week pairs from May 3 to October 4, 2026, the share of brands recommended again was higher than the share of URLs cited again. On GPT-5.6 Terra the weekly brand share ran from 75.1% to 81.4%, and the URL share from 42.0% to 46.7%.
## The top 10 held for a month
Of the brands in a category's top 10 in the weeks of July 26 to September 6, 2026, 89.0% (631 of 709) were still in that category's top 10 four weeks later. By category, the share ran from 81.4% in GEO to 100% in Email Marketing. In the GPT-5.4 weeks of May 3 to June 21, it was 83.2% (686 of 825).
## One answer reshuffles more than the category does
Inside a single question, the list of recommended brands moves more. For each question asked in two consecutive weeks on GPT-5.6 Terra, we compared the brands its two answers recommended. Across 7,132 such pairs, the median overlap was 0.571, counting brands named in both weeks against brands named in either. Only 724 pairs (10.2%) recommended exactly the same brands both weeks. An answer named 4.88 brands on average.
Counted at category level over the same weeks, the share of recommended brands that were recommended again the next week ran from 75.1% to 81.4%.
## Where recommendations move too
| Category | Brands recommended again next week | URLs cited again next week | Top 10 still top 10 after four weeks |
| - | - | - | - |
| CRM | 80.0% | 39.6% | 83.1% |
| Marketing Automation | 87.5% | 44.4% | 97.2% |
| Email Marketing | 85.9% | 47.0% | 100% |
| Project Management | 85.2% | 41.7% | 86.3% |
| Help Desk | 82.0% | 45.4% | 91.4% |
| Influencer Marketing Platforms | 71.4% | 51.9% | 94.4% |
| AI Writing Assistants | 69.6% | 42.0% | 84.5% |
| AI Image Generators | 83.7% | 43.3% | 84.5% |
| SEO | 86.2% | 44.2% | 87.1% |
| GEO | 50.5% | 41.4% | 81.4% |
*DecaGEO, ChatGPT (GPT-5.6 Terra), US, English. Week pairs from July 26 to October 4, 2026. The top 10 column covers starting weeks July 26 to September 6. Definitions in [How we measured](#how-we-measured).*
GEO is the exception. Its brands were recommended again the next week 50.5% of the time (205 of 406), close to its URL share of 41.4%, and 47.6% three weeks out. In none of its ten week pairs did GEO keep more than 64.1% of its brands. Influencer Marketing Platforms (71.4%) and AI Writing Assistants (69.6%) came next. The other seven categories kept between 80.0% and 87.5% of their brands over the ten week pairs. Even in GEO, 81.4% of the top 10 were still in the top 10 four weeks later.
The model change week moved both counts. From July 19, measured on GPT-5.4, to July 26, measured on GPT-5.6 Terra, 30.4% of cited URLs and 63.6% of recommended brands carried over, the lowest values on each line. [Did your ChatGPT ranking drop because of the model?](/docs/rank-drop-me-or-the-model) covers how to read a week like that.
## What this means for publishing to stay recommended
DecaGEO's panel measures how long a recommendation lasted while the questions and the model stayed the same. It counts recommendations and citations, not what each brand published, so whether a new page or a refresh keeps your own brand recommended is something to watch on your brand over the weeks after you ship it.
In every GPT-5.6 Terra week, more than half of the URLs cited the week before were not cited again, while at least three in four recommended brands were recommended again. A page dropping out of ChatGPT's citations happens to most cited pages every week. On its own it says little about whether the brand lost its recommendation. Base a refresh or a report on the brand's recommendation numbers over several weeks, and treat the citation list as a view of which pages ChatGPT used that week. For the kinds of pages it cites, see [Which pages does ChatGPT cite?](/docs/which-pages-chatgpt-cites)
## What to check when your URL drops out of the citations
1. **Open your category's ranking.** Without signup, it shows the latest week's rank, weekly movement and [Mention Rate](/docs/metrics/mention-rate) for every ranked brand. Mention Rate is the share of the category's answers that recommend the brand. Once you sign in, including on the free plan, switch the week to set your Mention Rate beside last week's. If it held, ChatGPT recommended you as often as the week before, whatever pages it cited. Rank follows DECA Score, which is measured against the week's top brand in the category, so your rank can move when other brands move. Mention Rate shows whether you yourself were recommended more or less often.
2. **Check whether the third-party sources around you stayed.** In the [Off-Page Citations](/docs/using/off-page-citations) tab, also open once you sign in, the top line counts how many of the category's answers mentioned your product that week. Below it are the reviews, forums, references and directories cited in the category, with the exact URLs ChatGPT pulled from and whether your product is mentioned there. Switch to the week before to see whether the sources that mention you were still cited. The tab lists third-party sites only, so your own pages do not appear in it.
3. **If you noticed it on one question, read that question's Brand Ranking Trail.** In [Prompt & AI Response](/docs/using/prompt-and-ai-response), also open once you sign in, the Brand Ranking Trail lists every brand an answer recommended, in order. Compare this week's trail with last week's. In the panel, only 10.2% of question pairs recommended the same brands two weeks running.
4. **Wait for more than one week before calling it a loss.** [Chart History](/docs/using/chart-history) on the free plan covers the last four weeks, and you can track one product with alerts through [Add Your Brand](/docs/using/add-your-brand). Pro keeps the full chart history and adds weekly change alerts for unlimited tracked products.
5. **Report the two numbers in separate sentences.** For example: "One of our pages dropped out of ChatGPT's citations this week. ChatGPT recommended us in the same share of category answers as last week."
## FAQ
### Do ChatGPT brand recommendations fade as fast as AI citations?
Not in DecaGEO's ChatGPT panel. From July 26 to October 4, 2026, on GPT-5.6 Terra, 44.0% of the URLs cited in a category one week were cited again the next week, while 78.3% of recommended brands were recommended again. Three weeks out the shares were 41.3% and 77.7%. See [the weekly comparison](#citations-turn-over-every-week-and-most-recommendations-carry-over).
### If ChatGPT stops citing my page, did it stop recommending my brand?
Not necessarily. In every GPT-5.6 Terra week, more than half of the previous week's cited URLs were not cited again, while at least three in four recommended brands were recommended again. Check your Mention Rate on the category ranking before reporting a loss. See [What to check](#what-to-check-when-your-url-drops-out-of-the-citations).
### Do I need to keep publishing to stay recommended by ChatGPT?
Most recommendations in the panel carried over from week to week, and 89.0% of top-10 brands across the ten categories were still top 10 four weeks later. The panel counts recommendations, not publishing, so judge your own publishing by your Mention Rate in the weeks after a page goes live. See [What this means for publishing](#what-this-means-for-publishing-to-stay-recommended).
### Why did my brand disappear from one ChatGPT answer but stay in the ranking?
A single question's brand list changes more than the category's. Only 10.2% of question pairs recommended exactly the same brands in two consecutive weeks, and the median overlap was 0.571. At category level, 75.1% to 81.4% of recommended brands were recommended again each week. See [One answer reshuffles more](#one-answer-reshuffles-more-than-the-category-does).
### Which categories have the least stable ChatGPT recommendations?
GEO, where 50.5% of brands were recommended again the next week. Influencer Marketing Platforms (71.4%) and AI Writing Assistants (69.6%) came next, and the other seven categories ranged from 80.0% to 87.5%. See [Where recommendations move too](#where-recommendations-move-too).
## 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.** Weekly rankings for the weeks of May 3 to October 4, 2026, ten categories. The model changed once, from GPT-5.4 to GPT-5.6 Terra in the week of July 26. Week pairs are grouped by the models of both weeks. GPT-5.6 Terra pairs start in the weeks of July 26 to September 27, GPT-5.4 pairs in May 3 to July 12, and the model change pair is July 19 to July 26.
* **URLs cited again.** A cited URL counts once per category and week, after lowercasing and removing the query string, the fragment and a trailing slash. It counts as cited again if the same URL is cited in the same category in the later week.
* **Brands recommended again.** A brand counts if it is on the category's ranking that week with at least one recommendation. It counts as recommended again if it is on the same category's ranking with at least one recommendation in the later week. A brand that ChatGPT still named but that fell off the ranking counts as not recommended again, which pulls the brand share down rather than up.
* **Top 10.** Brands ranked 10th or better in a category. Brands with equal DECA Score share a rank, so some weeks have more than ten. "Still top 10" means ranked 10th or better in the same category four weeks later.
* **Overlap within a question.** For each question asked in two consecutive GPT-5.6 Terra weeks, the brands recommended in both answers divided by the brands recommended in either, from July 26 to October 4, 2026.
Related: [Mention Rate](/docs/metrics/mention-rate) · [Reading Chart History](/docs/using/chart-history) · [Did your ChatGPT ranking drop because of the model?](/docs/rank-drop-me-or-the-model)
This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.