First-party evidence is material that only your business can publish because it comes from your own work: your data, your test results, your screenshots, your project outcomes. Original research is a structured form of it, with a defined question, a method and a dataset. Both give search engines and AI assistants something they cannot find elsewhere, which makes your page the natural source to cite.
Why it works
When an AI assistant combines several sources into one answer, a fact found on many sites can be credited to any of them. A fact found only on your site can only be credited to you. The 2023 GEO research paper by Aggarwal and colleagues found that adding statistics and quotations from sources raised visibility in AI-generated answers by up to about 40% in their tests. Google’s guidance on helpful content asks the same thing from the other side: does the content provide original information, research or analysis?
It also supports the “experience” part of E-E-A-T. A screenshot of a real crawl report or a table of real before-and-after numbers shows that the author has done the work.
Types of first-party evidence
| Type | Example for a services business |
|---|---|
| Aggregated project data | Median page speed improvement across the last 20 sites you optimised |
| Case results | A client’s featured snippets before and after a rewrite, with permission |
| Tests | What happened when you changed one variable on your own site |
| Screenshots and logs | A server log extract showing which AI crawlers visited |
| Survey data | Answers from your clients or prospects to a set of questions |
| Process documentation | Your real checklist, with the reasons behind each step |
| Pricing information | Typical price ranges and what drives them |
How to publish it properly
- State the method: where the data came from, how many cases, what time period.
- Give the numbers exactly, with units, next to the claim they support.
- Say what the data does not show, such as small samples or a single industry.
- Get written permission before naming a client, or anonymise the data.
- Keep the raw material, so you can answer questions or update the figures later.
- Put the finding at the start of its section, so it is the passage that gets retrieved.
What to avoid
- Invented or rounded-up figures. When someone checks, trust is lost, and an AI assistant repeating a wrong number with your name attached is worse than no citation.
- Screenshots with client data visible without permission.
- Presenting one case as a general rule.
- Quoting other people’s statistics as if they were your research.
Starting small
Original research does not need a large budget. A services business already has data in its projects, support tickets and sales calls. One well-documented finding, such as the most common technical problem in the last fifty audits and how often it appeared, is more useful than a long page of general advice.
This site follows the same idea on a small scale. The home page shows the live llms.txt file the site generates, and the guide on entity resolution describes a real naming inconsistency found while building the site.
Frequently asked questions
Is first-party evidence the same as a case study?
A case study is one format for it. First-party evidence can also be a single figure, a screenshot, a test result or a documented process inside any page.
Can I use data from tools like Search Console?
Yes, for your own sites or with client permission. Tool data becomes first-party evidence when you add what it shows and what you did about it.