Which Chicago agencies do AI assistants name?

Each number is the count of that engine's 150 answers in which the firm was named, once per answer. Naming a firm twice in one answer still counts once.

Gemini and Claude are counted through their developer APIs; Google AI Mode through a SERP vendor's capture of the live AI Mode surface. None of the three is a signed-in, personalised session. These counts tell you who an engine reaches for; they are not a measured share of what any one person sees.

A count out of 150 runs carries about ±8 points of uncertainty at the halfway mark — a standard 95% confidence interval derived from the run count, not separately measured — so treat small gaps between neighbouring firms as a tie.

No position or rank is reported. The order an answer lists firms in barely repeats between runs, so ordering them would be inventing a signal.

Only 4 firms appear in all three engines' top ten: Flamingo Agency, Ice Nine Online, Qoulomb, Thrive Internet Marketing Agency. Of the 75 firms in the table above, 29 were named by one engine only — invisible on the other two across 300 answers. If you are checking your own visibility on a single assistant, you are reading one of three different markets.

Each engine has a favourite, and they are not the same firm. Gemini named Digital Third Coast in 105 of 150 answers; Google AI Mode named Comrade Digital Marketing in 53 of 150 answers; Claude named Flamingo Agency in 62 of 150 answers.

Flamingo Agency was named most often overall — 158 of 450 answers, across all three engines. Breadth across engines and depth on one engine are clearly different achievements.

Questions. 5 city-named buyer questions about AI visibility agencies in Chicago. City-named, never "near me": named-city questions repeat far more stably.

Runs. 150 answers per engine, spread over 5 calendar days, 2026-09-09 to 2026-09-13, America/Chicago. Zero runs errored. A single day is a snapshot, not a baseline, so nothing here rests on one day's answers.

Engines. Gemini — API with search grounding; Google AI Mode — the live AI Mode surface, pulled through a SERP vendor; Claude — API with web search. ChatGPT is absent on purpose: I run it by hand in the app for clients, and hand-run answers are too few to belong in a table like this.

Extraction. Every answer is read by a separate model that lists the businesses named, blind to any list of firms. Nothing is matched against a target.

Name variants. Names are reduced to their distinctive words, and one form counts as another when its distinctive words are a prefix of the other's, so "Comrade" and "Comrade Digital Marketing Agency" are one firm. The 23 merged groups are listed in full below.

What was left out. 45 named entities are not Chicago agencies and are excluded: the assistants themselves, listing platforms, software products, news publications, trade bodies and directories. Every one is named below with its reason. Where a name was ambiguous I kept it on the list rather than delete it.

Re-derivable. Every count on this page comes out of a run log, not a spreadsheet. The build script and its checks re-derive each number from the raw answers and refuse to publish on a mismatch.

Moderact is a Chicago agency that counts how often AI assistants name a business — the same buyer questions, run many times across days — and fixes what the count shows.

The same method runs on your questions and your city: your count per engine, who else gets named for the questions your buyers ask, and the sources cited while the answer is written — then the fix list, then the count again.