Ask an AI assistant for the best moisturizer, a good dog food or marketing software for a small team, and you get a shortlist. If your business isn't on it, the buyer never hears of you.
Those lists change when the model behind the assistant changes. A model update can take you off the list without anything changing on your side. So we built a way to watch them: the Runner AI Shortlist Index.
What we asked
We wrote 72 buyer questions in 12 categories, from sunscreen and pet insurance to sneaker resale sites, shipping software and agriculture news. None of them names a brand.
We put every question to ChatGPT, Claude, Gemini and Grok, on each one's current model and on the models before it, on the same day. Then we counted the brands each answer recommended.
What changed
- ChatGPT's list got shorter. GPT-6.1 Sol named 3.5 brands per answer. GPT-5.6 Sol named 4.5 on the same questions, and its list was shorter on 38 of the 72. The everyday model, GPT-6.1 Terra, went from 4.4 to 4.0.
- Claude names the most. Claude Sonnet 5.5 named 9.2 brands per answer, up from 8.6 on Sonnet 5, about twice what ChatGPT names.
- Grok grew, mostly from search. Grok 4.7 named 5.9, up from 3.8 on Grok 4.3. It looked things up on the web for 97% of the questions. Grok 4.3 did for 63%.
- Gemini sits in the middle at 5.0, answering from its own knowledge.
Why it matters
On ChatGPT a shortlist is now about four names long. When an answer names four brands, the fifth isn't lower on the page. It isn't on the page at all.
And each assistant builds its list its own way. A brand that makes Claude's nine can still miss ChatGPT's four.
A model update can take you off the list without anything changing on your side.
What we do with it
We ask the same questions every week. When we move to an assistant's newer model, we keep asking the old one beside it for four weeks, so a change shows up the week it happens. Every answer is kept, with the sites it cited.
Runner customers see it in the app, next to their own AI visibility check: how many brands each assistant names this week, what changed with each model update, and the sites AI answers lean on in each category.
How we measured
Answers came through each vendor's API on November 12, 2026, with web search on where Runner can turn it on (ChatGPT, Claude and Grok). A brand counted only when the answer itself named it. This week is one answer per question per model, so a single category on its own is a small sample. Every weekly run adds to it.
If you want an AI team that checks where you show up and works on the answers you're missing, build your team (opens in a new tab). Plans come with a 14-day money-back guarantee.
Give your business a growth team.
Create your account and pick a plan. Your team is built once the plan is paid.
