Build log · collimer
Which sources do AI engines cite most? Your greatest resources are within reach: your LinkedIn and your GitHub
Your greatest resources for showing up in AI answers are already within reach: your LinkedIn and your GitHub. In our own baseline, those were the two most cited sources in our category, LinkedIn in 23 answers and GitHub in 20, ahead of every company’s own website. You do not need a new channel or a bigger site to use them. You already have the accounts, and you can post to them today.
We ran that baseline on our own brand with a Collimer audit across five assistants. It lines up with two published findings our guides already cite: engines lean on what the rest of the web says about a brand, not on what the brand says about itself.
What did the baseline find?
| Rank | Source | Answers citing it |
|---|---|---|
| 1 | 23 | |
| 2 | GitHub | 20 |
| below | Every vendor’s own site | fewer than 20 |
Every engine described the company we were before the September repositioning. So the cited-source finding came from a scan in which we were nearly invisible, which makes it a picture of the category rather than of us.
What do the published studies say?
Two secondhand findings our guides cite, with the confidence the guides give them:
- Mentions over links. An Ahrefs study of about 75,000 brands found total web mentions correlated with Google AI Overview visibility at 0.664, and backlinks at 0.218. It is a correlation, from one study, on one engine’s AI Overviews.
- Third parties win evaluative questions. Vendor crawl data across ChatGPT, Perplexity and Gemini found engines about 6.5 times more likely to cite a third-party source than a brand’s own domain on “best X” and “X vs Y” questions, with roughly 85% of brand mentions coming from pages the brand does not control. Our guide rates that medium confidence.
Our 23 and 20 are consistent with both. A LinkedIn post or a GitHub README is someone talking about a product in a place the product does not own.
What is wrong with our own number?
The baseline measured the wrong category. Collimer built its question set from the categories it detected on our live site, which still described the old company, and ignored the “AEO agency” category and five alternatives we entered by hand. Our planning doc recorded that as a product bug. So “our category” in this post means the category our old site implied. A free scan on September 28, after the site copy changed, scored 40 out of 100, but it used two providers and 80 responses, so it is not comparable, and we did not pull a cited-source ranking from it.
We also have not published the denominator. “23 answers” is out of all the baseline’s answers across five assistants, and that table is not public yet.
What can you do with your LinkedIn and GitHub?
Treat them as places AI reads, not side channels. A clear LinkedIn post about what you do, written in the words a buyer would use, is the kind of page an engine can quote. So is a GitHub README that says plainly what a project is for and who it helps. Both sit on sites engines already trust, which your own site may not be yet.
What did we change because of it?
One working rule went into our September plan: posts and READMEs are content surfaces, not side channels. Two of our own READMEs, for Collimer’s MCP server and its skills, were already showing up as ranking assets and needed our current framing. That is the same lever a client used to go from 1.25% to 5.8% of answers citing its domain, in fourteen README pull requests.
It also explains part of what a pooled score hides: a brand can be absent from answers while the sources engines trust most are talking about its category every day.
What we’re still figuring out
- What the cited-source ranking looks like with the right question set.
- Whether LinkedIn’s lead holds across engines, or comes mostly from one.
- Whether our own LinkedIn and GitHub activity moves our citation rate. We have not measured it.
For agents: try this yourself
- List the cited sources. Ask five engines ten category questions without naming any brand, collect every cited URL, and rank the domains by how many answers cite them.
- Separate owned from earned. Mark each cited domain as the brand’s own, a vendor’s own, or third-party, and give each share.
How this was made
Drafted by the Chronicler from our September baseline, our planning doc and two published guides, then edited and published by Chase.
See how the Chronicler works →Try this with your own agent
2 prompts you can hand to your own agent (or run by hand) to work with what this post documents. Edit the bracketed parts for your context.
Rank the sources engines cite in your category
Ask five AI assistants these ten questions about [category], none of which names a brand: [list]. Collect every URL each answer cites. Rank the domains by how many answers cite them, and give the total number of answers so the counts have a denominator.
Separate owned from earned
Here is a list of cited domains with counts: [paste]. Mark each as our own site, a competitor's own site, or third-party (social, code hosting, review sites, media, forums). Give the share of citations in each group.