Thoth transparency post for May 2026
It's mid-May, so time for another update on Thoth's ongoing activities.
Thoth Overview
Thoth is an AI curator who scans the Steem blockchain for posts that bring lasting value to the Steem blockchain. When found, it writes a set of summary posts and replies, with beneficiary settings for any/all of four recipient account types: itself (currently disabled in testing); the authors of the posts that it finds; an influence-weighted random selection of delegators; and to the null account, for burning. It is currently running twice per day under the @thoth.test account.
The content that Thoth supports can include active content and also archived content that has already passed its payout time.
Through this mechanism, Thoth makes two things possible that were not happening before: (1.) the blockchain can deliver fully passive rewards to delegating accounts; and (2.) The blockchain can direct rewards to authors for posts that already paid out - perhaps even months or years ago.
To be clear, reward distribution amounts are entirely determined by blockchain voting consensus. Thoth is not capable of guaranteeing any specific rate of return to delegator or author accounts.
In order to be a good AI Steemizen, Thoth complies with the suggestions in this draft netiquette proposal for Steem automatons (until a community agreed set of guidelines becomes available for automated content creators).
Recent code updates (New)
Over time as available models changed, the LLM prompts have slowly gotten out of step with the models that were available, so last weekend I updated the prompts in an effort to get (roughly) consistent results from 3 different free-tier models that Thoth is capable of using: gemma-4-31b-it and gemini-3.1-flash-lite-preview from Google, and Qwen-3.5-27B-Derestricted from ArliAI.
These updated prompts are currently in testing, both in Thoth itself, and also through skills that I have implemented for myself in Brave Leo.
Additionally, I updated some of the API parameters to values that are (hopefully) better suited for today's models. Finally, some changes were made last month to try to increase the variety of authors whose posts get selected.
How selective is Thoth? (New)
I recently added a new section of stats at the end of each Thoth run so I can get an idea of how many posts are being passed through screening, and how many posts are being rejected by AI curation. So, now we can get a general idea about how picky Thoth is when selecting content.
Basically, because the test account implements a filter for inactivity time, there are two general "tiers" of selectivity. The existence of inactive accounts means that older posts are much more likely to be skipped over than newer ones. So, I'll show two sets of numbers here:
Older posts
Here are some numbers from this morning's run, which focused on posts from 2018:
Total Posts Evaluated: 15769
Total Posts Accepted for Curation: 7
Total Posts Rejected: 15762
Acceptance Rate: 0.04%
In short, only 0.04% of posts (7 out of 15,769) made it past the screening phase, and of those 71% (5 of 7) were accepted by the LLM.
Newer posts
Similarly, here are the numbers from a run yesterday with posts from 2025:
Total Posts Evaluated: 1456
Total Posts Accepted for Curation: 13
Total Posts Rejected: 1443
Acceptance Rate: 0.89%
So, when looking at recent posts, the screening accepted 0.89% (13 out of 1,456) posts and the LLM accepted 38% (5 of 13).
I don't have the ability to aggregate statistics across posts (yet), but I'd estimate that these numbers are fairly typical for the two eras. They're also somewhat consistent with the previous version of LLM prompts (but maybe a little bit more restrictive).
Now, let's get on to the blockchain's rewards distributions.
Here are summary tables with estimates:
Summary of PowerBI visuals
| All time | 30 days | |
|---|---|---|
| Total beneficiary payouts (STEEM * 2 ) | 11,847 STEEM | 2,087 STEEM |
| Burned beneficiary rewards ( STEEM * 2 ) | 2,988 STEEM | 522 STEEM |
| Author beneficiary rewards ( STEEM * 2 ) | 3,862 STEEM | 640 STEEM |
| Delegator beneficiary rewards ( STEEM * 2 ) | 4,996 STEEM | 914 STEEM |
| Number of unique non-delegating author beneficiary accounts | 887 | 131 |
| Number of unique delegator beneficiary accounts | 19 | 13 |
@thoth.test info from SteemWorld
| All time / Current | 30 days / Comment | |
|---|---|---|
| Curation Rewards | 3,845 | 732 |
| Total Delegated SP | 122,988 | n/a |
| Owned SP or Non-beneficiary delegated SP | 41,355 | These delegating accounts opted out of FAIRD beneficiary selection. |
| SP from beneficiary-eligible delegators | 87,467 | These delegators are included in FAIRD beneficiary selection. |
And here are the updated visuals
(Reminder that April is the last complete month, so the numbers for May are not complete yet.)









