What a CiteMe result can and can’t tell you

Branded versus non-branded questions

Most CiteMe questions ask what your buyer would ask, without your company name. Buyers who haven’t heard of you can’t ask about you by name, and the buyer’s question tests whether AI would recommend you unprompted. The Pro run also includes alternative-provider questions that name your company explicitly, to show how engines position you against alternatives when a buyer is already comparing.

How questions are written

The free snapshot uses the question you write, after a conservative cleanup: whitespace and formatting are normalized, filler openers like “please” are removed, and your company name and domain are stripped so the test stays non-branded. If location matters, the place is added to the question. The Pro run asks 105 buyer questions: 30 best-fit, 30 alternative-provider, 30 use-case, and 15 community questions. Most are written by an AI model from your website content, industry, company type, and location, then automatically validated and balanced; the community questions come from fixed templates, and the model-written groups fall back to templates if generation fails. Question generation is automated; no human reviews the questions before a run.

Engines

Questions run through three engines, each with live web search available. ChatGPT: the gpt-5-search-api model, built to search the web before answering. Gemini: the gemini-3.6-flash model with Google Search grounding enabled — the model decides question by question whether to search. Perplexity: the sonar-pro model, which is search-native — searching the live web is how it always answers. Sources are recorded as each engine reports them: pages an engine reports fetching are recorded as fetched sources, while a URL that only appears in an answer’s text is recorded as answer text, not as proof the page was fetched. Answers reflect exactly those configurations.

Sampling

The free snapshot takes one answer per engine. The Pro run asks each question 3 times per engine. A single answer can change from run to run, so repeated samples show whether a pattern holds or was a one-off.

What “named” means

In the free snapshot, “named” means the answer text literally contains the company name, a known alias, or the website domain — a plain text match, with a token-level fallback that catches the company’s distinctive name words appearing together (or, for one-word names, the word alongside the industry). In the Pro run’s answer analysis, a separate AI judge model reads each answer and reports whether the brand was mentioned, along with its position and the sentiment of the mention. Neither mechanism measures how often a model “really” recommends a firm, and we don’t publish detection-rate figures.

Sources

CiteMe records the source URLs each engine visibly shows with its answer. A visible source shows what appeared alongside the answer. It doesn’t prove the engine’s reasoning, and changing a source doesn’t guarantee a different future answer.

Variation

Answers change with the engine, the model, the wording, the location, and the date. Every CiteMe result carries its run date, and a result describes that run — not a permanent state.

What we will never claim

• We don’t guarantee rankings, mentions, or recommendations. No one controls what a model says.
• One answer is not a ranking. Every result shows the question, engines, date, and sample size.
• A visible source is a clue, not proof of why an answer was given.

Showing other companies’ results

CiteMe has no affiliation with the companies shown in published results. Real results for named firms are published only with their written approval, or when the firm is a well-known public brand tested independently.

Corrections

If you believe a published result is wrong, email tyler@schoneman.io and we’ll review it and note any correction on the page.