When AI assistants mix up brand names, and how to fix it

An assistant that writes "Coop", "Coop Home Goods" and "Coop Sleep Goods" in different answers has not made three brands. But a counter that treats them as three will report three small shares instead of one larger one. Name confusion matters for two reasons: it distorts what you measure, and an answer may attach the wrong company or product line to a name. This post covers both, starting with how the Dopbot index handles it, because that is the part we can describe from the inside.

What does brand-name confusion look like?

There are three patterns, and they need different responses.

• A split. One brand appears under several strings: a short form, a longer legal or trading name, a product-line name, a spelling variant. Counting is split across them.
• A collision. Two companies share a name or a near-identical one, and an answer attaches the wrong one.
• A misattribution. A product is credited to a brand that does not make it, or the brand's name is attached to a different product line.

Splits are the easiest to find and fix in your own data. Collisions and misattributions need a person to read the answers and compare them with the facts.

How does the Dopbot index handle name variants?

The index counts a brand at most once per answer, and only when the answer recommends it. Names go through three stages, and each stage has a limit worth knowing.

1. Normalization. Capitalization, punctuation, accents and endings such as Inc. or Co. do not create separate brands. This is a rule applied to every name.
2. A hand-made merge list. For one brand that comes back under more than one name, a person adds a merge, and answers that use any of those names count as one brand under one shown name. The methodology gives two examples: Coop, Coop Home Goods and Coop Sleep Goods, and Technivorm, Technivorm Moccamaster and Moccamaster. As of October 2026 the list has 26 merges for 24 brands.
3. Review for the rest. Each month a person looks at the names in a category that start the same way. A variant nobody has reviewed yet is not merged. Where it is unclear whether two names are one brand, such as ASUS and ASUS ROG, they stay separate until a person decides. A program only suggests pairs to look at.

The lesson from building it: the cost of a wrong merge and the cost of a missed one are different. A wrong merge credits one company with another's recommendations. A missed merge understates a brand. We accept the missed merge until a person has looked, and we say so on the methodology page.

You can see the effect with a small made-up example. In an eight-answer category, a brand is named as "Alpha" in 3 answers and as "Alpha Home" in 2. Counted as two brands, it has 3 of 8 and 2 of 8. Merged, it has at most 5 of 8, and 4 of 8 if one answer names both, because a brand counts once per answer. These counts are an illustration, not data.

How do you audit your own name in answers?

You do not need a tool for a first pass.

1. Collect answers. Ask the five question types in the free manual method four times each on each assistant, in fresh sessions: 20 answers per assistant at the least.
2. Extract every string the answers use for your brand and its products, exactly as written.
3. Group the strings: exact name, short form, longer form, product-line names, parent company, misspellings, translations.
4. Count two ways: answers per string, and answers that use any string in the group (once per answer).
5. Check collisions. Search each string. Note other companies with the same or a near-identical name, and whether any answer attached their product or description to you.
6. Record by assistant. The same name can behave differently in different assistants, so keep the count per assistant, as the overlap post explains.

A sheet with one row per answer works: assistant, question type, string used, recommended or only mentioned, and a note on any wrong attribution.

What can you fix?

These are the documented requirements as of October 2026; check each page again before you change your data.

We found no published evidence that any step below changes how an assistant spells, merges or attributes your name. Each one removes an ambiguity that a system could otherwise copy.

• Use one primary name everywhere. Product titles, page titles, the About page, markup, feeds and retailer listings should lead with the same form. Put the full legal name where it belongs and keep it distinct from the trading name.
• Mark up the organization. Google's Organization markup has name, alternateName and legalName fields, and a sameAs field for pages on other websites with more information about you, such as a profile on a social or review site. Google says some properties help it disambiguate an organization from others, and that it does not guarantee that features using structured data will appear.
• Fill the brand field in product data. Google Merchant Center asks for the brand name that people shopping online generally recognize, and requires it for products with a clearly associated brand. It says submitting the product identifiers, GTIN and MPN, helps it understand what you are selling. OpenAI's product feed specification requires a brand field that matches the brand as shown on the product page, and has GTIN and MPN fields; an MPN should be submitted together with the brand.
• Check the retailer's brand field. On Shopify, the product's vendor field is described as the manufacturer, wholesaler or other vendor for the product, and a theme can show it. Whether your theme's markup uses it as the brand is something to check in your own page source. On marketplaces, find the brand field and align it.
• Look at public knowledge bases. Wikidata accepts an item that has a valid sitelink to a Wikimedia page, refers to a clearly identifiable entity described by serious public references, or meets a structural need. Google says an official representative of an entity in a knowledge panel can claim it and suggest changes, although not all panels are claimable.
• Separate lookalikes. If another company has a similar name, keep a short descriptor with your name in titles and structured data, and keep it consistent. This is a hypothesis, not a documented fix.

For the markup details, see the structured data guide and the feeds guide.

How do you know it worked?

Repeat the audit with the same questions and assistants after your changes, and compare the grouped count and the share of answers that use your primary name. With 20 answers per assistant a change of a few answers is within noise, so wait for a clear shift, as the post on how many answers to trust explains. Name changes may take time to appear in retailer systems and feeds, so re-measure after a few weeks. We have no source for how long that takes.

Frequently asked questions

How do I report a wrong or split brand name in the Dopbot index?

The methodology page asks you to email the category and what you expected to see, and it helps to give the date of the update. Every report is checked, and wrong names and mappings are corrected; brands are not removed or reordered on request.

Can I get a brand name corrected inside an assistant?

We found no documented route that is specific to brand names. OpenAI's help center describes thumbs up or down and reporting incorrect product information for ChatGPT shopping results, and Perplexity's describes a flag icon below an answer; neither is specific to brand names, so send the exact prompt, the answer and the correct name.