AI Visibility Index · August 2026

LLM observability (tracing, evaluation and monitoring for LLM/agent apps): who the AI engines actually name

As of August 21, 2026, across 47 AI answers (ChatGPT · Google AI Overviews · Google AI Mode · Gemini) to 4 buyer prompts, Langfuse was named most often (46 mentions). 13 vendors were named at all; only 4 by every engine that answered. Of the 427 sources the engines cited, 33% were vendor-owned pages.

Category

LLM observability (tracing, evaluation and monitoring for LLM/agent apps)

Measured

2026-08-21

Engines

ChatGPT · Google AI Overviews · Google AI Mode · Gemini

Buyer prompts

4

Answers

47

Raw answers

read the archive · download JSON

Source

CrediGeo measured AI visibility for LLM observability (tracing, evaluation and monitoring for LLM/agent apps) in 2026. Across 47 AI answers recorded on 2026-08-21, Langfuse was named most often (46 mentions).

Google AI Overviews appeared in 3 of 8 checks, for the other checks, a normal Google search in this category showed no AI answer at all. Its shares below are of the answers that appeared.

Vendors named in at least one answer, sorted by total mentions. Engines are never combined into one number. Stability counts flips, a prompt naming the vendor in some reruns but not others, across each vendor’s checks:

Mentions, not quality. This table answers “who is on the AI shortlist”, never “who is best.” Cells read “named / answers shown” per engine. Vendors not named in any answer are not listed, see the FAQ for what that does and doesn’t mean. Product aliases and sub-brands are matched (the alias set ships in the raw data). Vendor names that link out have a profile of their own, showing every category we measured them in.

Flips are counted against what sampling alone would produce at each vendor’s own hit rate. Across this table, 34 flips were observed where chance predicts about 67.9. Reruns are worth doing because a single ask gives you one draw rather than an answer. That is a different claim from saying this category is unusually unstable, and on this measurement it isn’t.

CrediGeo · measured 2026-08-21 · 4 prompts · ChatGPT ×20 · AI Overviews ×3 · AI Mode ×12 · Gemini ×12 · United States

The domains the engines cited while building these answers. Each is labeled by who owns it, because a “ranking” cited from a vendor’s own site is a different kind of evidence than an independent one. 33% of all 427 citations in this category came from vendor-owned pages:

Ownership rule: a domain is “vendor-owned” when it belongs to a vendor in this category’s alias set; review platforms and community/video surfaces are labeled as such; everything else is “other.” We label. We don’t exclude: the raw mention counts above include all sources.

What 47 answers can, and can’t, tell you

Enough to separate consistently-named from absent from flipping on August 21, 2026; not enough to make small share differences a ranking, one run moves a vendor’s number noticeably. AI answers move continuously. This page is a dated snapshot, re-measured monthly, not a permanent verdict. We sell the measurement discipline, not placements, and we make no promise about who appears here.

What is AI visibility for LLM observability (tracing, evaluation and monitoring for LLM/agent apps)?

AI visibility is how often an AI assistant names a product when buyers ask which LLM observability (tracing, evaluation and monitoring for LLM/agent apps) to use. When ChatGPT, Google's AI or Gemini answers "best LLM observability (tracing, evaluation and monitoring for LLM/agent apps)", it names a short list of specific vendors. This page measures who actually appears on that list, answer by answer.

How is this measured?

We ran 4 real buyer prompts repeatedly on each engine (ChatGPT · Google AI Overviews · Google AI Mode · Gemini) on August 21, 2026, in fresh sessions, and counted which vendors each answer named, 47 answers in total. Every raw answer is published and the method is public.

Why don’t the engines agree?

They read different sources, studies show different AI engines rarely cite the same pages. That per-engine split is the finding: of the 13 vendors named at all here, only 4 were named by every engine that answered. A single blended score would hide exactly that.

Why does Google AI Overviews have fewer answers than the other engines?

AI Overviews only appears for some queries. In this measurement it appeared in 3 of 8 checks. That trigger rate is itself part of the data: it tells you how often buyers even see an AI answer on a normal Google search in this category.

Is this a quality ranking?

No. This measures MENTIONS, how often an engine names a vendor, not product quality. A great product can be rarely named and a mediocre one often. It answers "who is winning the AI shortlist", not "who should you buy".

A vendor is missing, what does that mean?

Only vendors named in at least one of our 47 answers on August 21, 2026 for these 4 prompts appear here. Not appearing means our runs didn’t name them on that date with these prompts. Nothing more. If it’s your product, request your own run-by-run breakdown (free) and we’ll show you exactly what we asked and what came back.