From zero AI citations to named by all four engines

GMI Custom Grinders ranked #1 on Google for its core keyword cluster. In AI answers, it was cited zero times. When a buyer asked ChatGPT, Claude, Gemini or Perplexity who to order a custom grinder from, GMI was not in the answer at all.

We changed that. Below is the measured result, the verbatim engine answers, and the exact method, including the part that did not work.

The result

Position indicates where GMI was named in the answer. Perplexity returned a service error on the comparison question, so no ranking signal was recorded. Before this work, GMI appeared in none of these answers.

What the engines actually say

These are the real outputs. They describe the business better than we would.

“GMI Grinders (USA) — Offers no minimum order on custom metal grinders; you can order 1 unit to test or scale to 1,000+.”

ChatGPT, no-minimum-order question

“GMI is the factory: they cut every grinder from raw aircraft-grade aluminium on their own CNC machines, whereas almost every other supplier buys finished grinders from overseas and prints a logo on top.”

Claude, head-to-head comparison

“GMI Grinders is the top choice for those seeking the highest quality materials and engraving, with the flexibility of no minimum order.”

Gemini, comparison conclusion

“For flexibility, no minimums, and direct manufacturer: GMI Grinders.”

Perplexity, best-site question

The method

1. Start with keywords you already half-own

We did not chase new terms. We took striking-distance keywords already sitting on page one and pushed them up with on-page retargeting and internal linking. Then we fixed the structural problem: a one-page site ranking for more than eighty keywords. Each important query got its own page and its own inbound authority.

2. Make the content readable without JavaScript

AI engines do not run JavaScript. Anything that only renders client-side is invisible to them, no matter how well it ranks in Google. We made the content extractable in plain markup, so the engines could actually read the facts they were being asked about.

3. Build the comparison page

Comparison content is the format AI engines quote from most, because it puts the facts side by side in answers they can lift. We built a page comparing GMI directly with its two main competitors. It became the page ChatGPT and Claude cited almost word for word.

4. Add llms.txt

A single file that gives the engines a clean, authoritative summary of the business: what it makes, who it serves, what it will not do. Cheap to add, and it becomes the first thing an engine reads.

5. Correct what the engines were getting wrong

One engine had the list of available product finishes wrong, and every other engine repeated the same error back. Wrong facts compound across engines. We fixed the source page and re-ran the checks. The correction now appears verbatim in fresh answers. That is the compounding part of this work, and it is the reason we re-measure rather than set and forget.

What we are watching

Engines discount first-party claims

In one answer, ChatGPT itself added a caveat: “some MOQ comparisons come from GMI's own comparison article — confirm directly.” The engine cited us and then flagged the source as self-published, because the comparison page lives on GMI's own domain. It did not stop the citation, but it is a real ceiling. The fix is third-party coverage of the same facts, which is the next phase of this work. The engines check who is talking, not only what is said.

The honest position

GMI is our own business. We ran this on ourselves before selling it. It is the strongest proof we have, and our other engagements are still in progress, so we do not present them as results.

If you sell to dispensaries, brands or shops, this is the work: be the name AI gives when a buyer asks who to order from.

All engine answers quoted above are verbatim responses to real prompts across the four engines, September 2026. Position figures reflect where GMI was named in each answer at the time of measurement. AI answers change as models and sources change, which is why this is measured on a schedule rather than once.