These are the raw, unedited notes I wrote while shaping this article. They are closer to a brain dump than a prompt library: my thinking, philosophy, corrections, and direction. The ideas are mine. AI helped turn them into a smoother, more readable essay.

Research footprint

Open the evidence artifacts ↗
24human prompts
7Codex sessions
29committed artifacts
98.8Mtokens processed

This article was researched over 2h 20m of wall-clock collaboration, from to . The measured token total includes 98.4M input tokens (94.6M served from cache) and 464K output tokens. Reasoning tokens (134K) are a subset of output, not an additional charge.

This is a provenance measure, not a quality score, cost estimate, or environmental-impact estimate. Token counts come from the Codex session logs for the main article session and its research agents. The committed artifacts let readers inspect what all that computation produced.

Prompt 1
read docs/experiments/content-mix-evergreen-ideas.md , my blog is struggling, so far we still don't have real audience and readers, so far we only have my own personal engineering articles and we are not getting enough traction, the evergree ideas was recommended to me, but i want to fully ground ourselves, lets first iterate on these ideas
Prompt 2
ground ourselvesm lookup evidences that these articles trully make sense, and how can i stay and be genuine and write these kinda articles, but at the same time don't feel like i am farming for engagement? I wanna stay true to myself, and don't want to feel like i am fake.
Prompt 3
which one should we write about first
Prompt 4
i like this one: ## 3. The Honest Guide to llms.txt & GEO (What Actually Works)
we will write the article and learn lots of things ourselves both at the same time
Prompt 5
capture all these iterative work and findings in the md doc as an artifact, and proceed with iterations
Prompt 6
you can have a scrathpad with all the golden sections, information, distilled insights, cross references with rationale and so on, before you write a final article
Prompt 7
fanout if needed
Prompt 8
btw make sure all the fanout work gets saved into artifacts in this research folder
Prompt 9
red teaming, needs a green teaming as a followup to adapt and be flexible
Prompt 10
iterate on the final article as we are making progress now
Prompt 11
fanout as needed
Prompt 12
this article whatever we are about to write, needs to be an evergreen article that brings a lot of traffic to our gkoreli.com blog
Prompt 13
does this article touch what is GEO and aspects of GEO besides the llms.txt? is it helpful for knowing a lot more around it complementary to llms.txt?
Prompt 14
with evidences and cross references with rationale, distilled sections, insights and bullets? And know that it is Aug 25 2026, we need the bleeding edge, the latest information, established patterns and latest trends, for all things that we are touching.
Prompt 15
fanout as needed
Prompt 16
can we add somewhere to the article similar to reading time, to showcase how much LLM tokens where spent to build this article? Do you think thats beneficial to the reader?
Prompt 17
or like even showcasing like how many reasearch sessions, and how much investigation artifacts (all of which will get commited in the repo) took it to finalize and distill into the final article, with full provenance and transparency, like this is the reason it is so meaningful to read my articles, it takes so much tokens, and time and effort to research, investigate and distill into the final article, it took so much effort, tokens and time, human-steering and all that, so that someone can have a fully evidence driven final article.
Prompt 18
how did you determine the totalTokens? we have this current active session and so many sub agents that we triggered during the fanout, did you account for all of those? also make sure to capture in the AGENTS.md or somewhere, a deterministic clean reusable rule, how to extract and fully with full trust and evidence, figure out these numbers honestly and deterministically
Prompt 19
make commits as you go, update the prompts.md as you go too
Prompt 20
'(): ZodObject<{ parent_thread_id: ZodString; agent_path: ZodOptional<ZodString>; }, $loose>' is deprecated.ts(6385)
schemas.d.cts(439, 9): The declaration was marked as deprecated here.
(method) ZodObject<{ parent_thread_id: ZodString; agent_path: ZodOptional<ZodString>; }, $strip>.passthrough(): z.ZodObject<{
parent_thread_id: z.ZodString;
agent_path: z.ZodOptional<z.ZodString>;
}, z.core.$loose>
@deprecated — Use z.looseObject() or .loose() instead.
Prompt 21
prepare the article to be published
Prompt 22
update the totalTokens right before the final publishing
Prompt 23
relevant prompts.md doesn't seem to be up to date
Prompt 24
ready to push? all good? want to take a look one last time? how to share this on my Twitter?