First 50,000 GPTs | Trends and Insight

Community
,
1

Some interesting trends are starting to emerge now that we have reached 50,000 Public GPTs.

  1. GPT Count is doubling every 18 days. We’ll like 1,000,000 GPTs by March 1st, perhaps sooner with the recent opening up of Pro subscriptions

Screen Shot 2023-12-14 at 3.56.46 PM

  1. GPTs are multi-lingual. 37% of GPT descriptions are not written in English.

  2. GPTs are multi-purpose. Check out the most common words found in the names and descriptions of the first 50,000.

expert: 5297
guide: 3612
assistant: 2764
offering: 1305
data: 1258
advisor: 1243
help: 1126
master: 1116
code: 1110
coach: 1069
create: 1047
all: 1044
friendly: 916
game: 888
business: 880
based: 879
insights: 833
providing: 804
personal: 793
creative: 793
advice: 787
tailored: 752
detailed: 747
about: 744
me: 743
creator: 741
content: 735
mentor: 727
analysis: 725
engaging: 725
personalized: 718
generator: 703
design: 697
using: 691
interactive: 679
seo: 674
product: 671
web: 654
pro: 652
gpts.tapgpts.com.: 652
image: 651
creating: 645
through: 639
advice.: 638
will: 626
advanced: 623
prompt: 619
marketing: 606
text: 602
best: 600
get: 584
tutor: 580
digital: 576
companion: 574
virtual: 573
find: 565
bot: 564
language: 561
learning: 559
crafting: 552
professional: 548
art: 547
social: 539
guiding: 538
market: 534
helper: 532
assistant.: 528
insights.: 521
tech: 508
guru: 501
creates: 500
provides: 499
guidance: 494
provide: 493
style: 489
financial: 483
ask: 482
media: 474
buddy: 472
images: 467
legal: 465
writing: 461
english: 453
wizard: 451
news: 435
coding: 433
tool: 431
questions: 430
craft: 427
health: 427
helps: 425
guides: 421
research: 421
designed: 420
global: 418
analyst: 415
assists: 403
focusing: 399
user: 398
guide.: 397

Data from GPT Search API

2 Likes
2

I think these numbers are low.

I had personally cataloged over 35,000 publicly accessible GPTs as of two weeks ago, before I pulled the plug on the project.

Word is that, early last week, the total number of custom GPTs created broke two-million.

I don’t think you can draw the conclusion that the GPT count is doing every 18-days based on what I’m seeing there.

Other than that, the common words seems pretty similar to what I have seen, though it’s obviously missing the most common foreign-language words. There should be at least a few Chinese words in the list for it to be actually accurate.

2 Likes
3

I saw this trend starting in plugins as well. Developers were moving more towards “Super Plugins”. Basically, plugins that could do multiple things. I suspect this could be a telling sign that ChatGPT’s plugin inference/usage sucks and people were trying to make a better seamless experience. Or, it could be people were trying to do everything in one app just because they thought it was cool or they didn’t have an easy straight forward way to interface with other plugins.

Either way, it seems there is a lot of focus on multi-skill agents and ChatGPT is so far incapable of delivering the experience/capabilities developers want or need. Otherwise, they’d just be making single skill agents.

We’ll see. It’s definitely a great sandbox/petri dish to watch.

1 Like
4

Wow that is amazing. Do you have a source for the 2 million number? Does that mean most are private?

Our database is close to 65,000 public currently but there is plenty of opportunity to be more comprehensive.

5

Yes, he does have a source for that number.

As for the quality of that number, it may just be the number of people that pressed “create GPT” and made a database entry, or other farmable statistic. Didn’t ask.

1 Like
6

Are they all unique? Have you checked for duplicates?
https://chat.openai.com/g/g-2fkFE8rbu and https://chat.openai.com/g/g-2fkFE8rbu-dall-e are for the same gpt, but on some resources both are present. There are thousands of examples.
I found a resource without duplicates and there are about 30k now. Maybe not full, but at least unique

7

As of about 3 weeks ago now I had identified over 35,000 unique GPTs.

The process was very straightforward, I used the Google, Bing, and DuckDuckGo search APIs.

Basically, I searched based on keyword + site: https://chat.openai.com/g/.

I started out with a list of, I think, the 500 most common English words, the digits 0–9, and all 26 letters. I searched all three sources, for all ~530 keywords, for as many pages as we available, there was obviously tons of overlap and duplicates. When I finished with those runs, I split the text in the name and description of each result into words and kept only the unique words from each GPT. Then I tallied how many GPTs used reach word.

Using this as my new keyword index, and removing any keywords which appeared in the initial list, I repeated the process with the top 2,000 new keywords.

Lather, rinse, repeat.

At some point I grabbed some word frequency lists for Spanish, French, German, Korean, Japanese, Chinese, Hindi, and Russian and repeated the process for those word lists.

Based on how long I was at it and how many URLs for custom GPTs I had found at that point, I would guess there were at least 50,000–60,000 custom GPTs indexed three weeks ago and there’s probably 150,000–300,000 that could be found right now.

The main reasons why I abandoned my project was due to time and OpenAI has very aggressive anti-bot settings with CloudFlare making verifying the GPTs were still active and available more cumbersome than I cared to deal with (and I didn’t want to jeopardize my OpenAI accounts by circumventing them). There was also the fact that I was bumping up hard against the free API limits.

I’m absolutely certain that if someone wanted to invest a couple hundred dollars into search API calls they would be able to, starting from zero, catalog over 100,000 unique GPTs in a couple of hours.

Even using just what’s freely available I’m sure someone could pull 50k+ in no time.

Like I said before, multiple millions of custom GPTs have been created and there are lots of them out there which have been shared.

The real feat would be to be able to verify they are still alive and available, a difficult task given the CloudFlare settings.

1 Like
8

Great question. Yes they are all unique. I removed the “slug” from the url and save each GPT by the gptid.

9

Great point on the “active”. There are a lot of lists out there that seem to inflate their “counts” by including inactive GPTs. We have checked each gpt to ensure it is and do a re-check every week to ensure.

Thanks in general for your advice on your approach. We now have DuckDuckGo integrated as well as Google and Bing Web Search APIs. This along with a few other sources is proving to be a more comprehensive approach.

Were up to 85,000 gpts and are starting to get good data on popularity and quality to include in the free API.

Any other advice from the community to continually improve the data is welcome.

1 Like