Survelium Docs
Docs menu

How we collect data

How we read an answer

How we find your brand and your competitors in an answer, and how we determine position, sources, tone and new brands.

Each answer is kept in full (you see it on the Responses page) and then analyzed. The analysis produces the mentions, position, sources, sentiment and new competitors.

Your brand

We search the answer text for the brand name, its variants and the domain (see Brands). Matching ignores case. The brand counts once per answer, however often it appears.

If the assistant writes your name differently from your settings (shortened, without "Inc.", with a different spelling), add that form to Variations. Otherwise the answer doesn't count as a mention.

Competitors

An AI model reads each answer and notes every business it names, with or without a website. It is told who you and your direct competitors are, with their aliases and sites, so "Axiotime" and "Axiotime Traduceri" are recognized as the same business. Addresses, buildings, malls and platforms (Google Maps, Facebook) are not businesses and are not noted. A name the model gives that does not appear in the answer is dropped.

Your direct competitors are also searched for directly, by name and aliases, as a whole word: "Equip" does not match in "equipment". Matching ignores case and diacritics. See Competitors.

Position

Position is the order of first appearance among all the businesses the answer names. The first one named is in position 1. Businesses you marked Ignored take no place. See Average position.

On ChatGPT, when the answer contains a list of local businesses (with a map), your position is your place in that list.

Sources

An answer's sources are the links the assistant cites, plus the addresses written directly in the text. For each one we keep the page, the domain, the order (1 = cited first) and whether it's your domain. Each cited page also gets a type (news, opinion, list, guide, review, comparison, directory…). See Used as source and Your domain share of citations.

New brands

A business named for the first time is added to the Competitors list as No relation (the Suggestions tab). It counts in share of voice and positions, among "others"; mark it Ignored if it is not a competitor, or direct to track it. Exception: on ChatGPT, a business that appears in the same local list as you is automatically marked direct. Automatic rules never change a brand you marked yourself.

A business is kept with a website only when that site is its own:

  • the site is among the answer's sources;
  • none of its cited pages is a directory, a list, a review, a comparison or an article;
  • it is not a platform (Google, Facebook, LinkedIn and the like);
  • the answer does not give the same site to two different businesses — that is the page it read them both off.

Otherwise the business is kept without one. You can add or correct a site on the Competitors page.

Sentiment and stance

For each brand named we determine:

  • tone (positive, neutral, mixed, negative) — for your brand, from the passage where it appears;
  • stance — the role the answer gives it: Recommended, Conditional, Listed, Cautioned, Comparison, Context.

An AI model that reads the answer does this. Details and formulas in Sentiment and stance.

What the answer says about each brand

For your brand and each competitor named, we also note what the answer says about it — "fast delivery", "expensive", "certified" — sorted into fixed categories, each as a strength, a weakness or a plain mention, with the quote. Details in Brand perception.

AI searches

Some assistants tell us what they searched for on the web before answering. We keep these with the answer, and you see them in AI searches.

When the analysis gets it wrong

Automatic analysis can miss an unusually written mention or score a tone wrong. The full answer is always on the Responses page, so you can check any number.

View as Markdown · llms.txt