Answer engine optimization

How to get cited by ChatGPT using Hermes

ChatGPT does not recall your brand from memory. It retrieves a handful of pages and writes its answer from them, so getting cited means getting onto those pages, in a form it can lift. With Hermes, the workflow is: find the buyer questions where ChatGPT names a competitor instead of you, look at which URLs it keeps citing for those questions, and publish a page that out-answers them - leading with the answer, structured as a table or list, and honest about where competitors genuinely win. Then re-measure a few weeks later, because publishing without re-measuring is guessing.

How ChatGPT actually decides who to name

When a question implies recency or a recommendation, ChatGPT searches the web and writes its answer from the handful of pages it retrieves. It is not recalling your brand from training data - it is reading a few pages, right then, and summarising them. So the question is not 'how do I get into the model', it is 'how do I get onto the pages it pulls'.

What moves the number on ChatGPT

1. 1Get named on the third-party pages it retrieves - the roundups, comparisons and 'best X for Y' listicles that already surface for your query. Being in someone else's list beats a page on your own domain, because the model treats an independent source as more trustworthy than a vendor claiming to be the best.
2. 2Answer the exact question in the first 100 words of your own page. The model lifts the passage that answers the question; if yours starts with a brand story, it lifts a competitor's.
3. 3Publish a comparison table that names your competitors honestly. Tables survive chunking intact and get quoted more than prose, and a page that surveys the field is the kind of source a model cites when asked to survey the field.

Doing it with Hermes

An open agent runtime from Nous Research, with a skill system that lets you hand the agent a playbook rather than a prompt. The choice when you want the loop to run unattended and you would rather install a capability than describe one.

1. 1Add the repo as a skill source and install: `hermes skills tap add gsmmediaro/fulcru`, then `hermes skills install fulcru`. Hermes scans third-party skills before installing and shows you the verdict, so you can read what you are about to run.
2. 2Register the Fulcru MCP server in your Hermes `config.yaml` under `mcp_servers`. The skill teaches the agent the playbook; the MCP server gives it the measured data to run the playbook against.
3. 3Ask the agent to audit your domain. It builds the buyer question set, runs each question, and records who gets named and which URLs keep appearing across answers. Those recurring URLs are the pages the engines actually read in your category.
4. 4Have it draft the page for the worst gap, then read the draft yourself before it goes anywhere near your domain. Publish, then let the agent re-run the same set on a schedule so you are comparing against a fixed baseline rather than a moving one.

What this will not do

The part everyone skips

Re-measuring. A published page that you never check is a belief, not a result. Run the same question set again a few weeks later and compare: were you named in 2 of 10 answers before, and 5 of 10 after? That number is the only thing that tells you the work landed - and when it does not move, that is information too. It usually means the answer set is dominated by a source you have not gotten onto yet.