Why AI recommends some brands—and ignores others

What can make a brand appear in an AI answer?

The answer depends on the question and the product answering it. A buyer asking for a shortlist of tools may receive named vendors, while a question about how a tool works may receive an explanation and links instead. ChatGPT Search, Google AI Overviews, and Google AI Mode do not present identical results. Google says its two AI search experiences can use different models and techniques, so their responses and links may vary.

No public documentation establishes a universal percentage weight for reviews, awards, backlinks, or mentions in AI brand recommendations. Treat claims that one factor explains a fixed share of recommendations as unverified unless they disclose the prompts, products, dates, sample, and calculation.

First check whether the page can be found

For a supporting link in Google AI Overviews or AI Mode, Google says a page must be indexed and eligible to appear in Search with a snippet. It also says there are no additional technical requirements specific to those AI features. Meeting the requirements is eligibility, not a promise of inclusion.

OpenAI advises publishers who want their content discoverable in ChatGPT Search to allow OAI-SearchBot. Its guidance separates search discovery from potential model training. Check the canonical URL, HTTP response, robots rules, noindex directives, and whether the important answer is visible in the page content. Use Search Console URL Inspection for Google’s view of a priority page.

Then check whether the page answers the buyer’s question

Compare the exact prompt with what the page actually explains. A product page that says “AI visibility platform” but never defines supported answer surfaces, collection method, prompt limits, or the buyer it serves leaves a reader with little evidence for a comparison. A vague page can be technically indexable and still be a poor answer for the question.

Write the missing answer for people first. State what the product does, what it measures, what it does not measure, and where a buyer can verify the claim. Google’s people-first content guidance asks whether a page offers original information and enough detail to help its audience complete a task. That is a useful editorial standard, though it does not guarantee an AI citation.

Separate brand mentions, recommendations, and citations

These signals answer different questions. A brand mention shows that the name appeared. A recommendation shows that the answer presented it as a fit for the stated need. A citation links to a source page. None of these alone proves a visit, a lead, or a sale. When a competitor appears, open the cited pages and record what they actually say before assuming a causal relationship.

The note identifies an editable gap and a way to test it. It does not claim that adding a phrase will force the model to recommend the brand. Relevant third-party coverage can give buyers independent context, but outreach should earn editorial consideration through useful data or expertise rather than promise a ranking outcome.

Measure a fixed set before judging a change

1. Choose questions from real discovery, comparison, and purchase tasks. Keep the wording and intended buyer visible.
2. Record the product and mode, market, language, date, answer text, named brands, recommendation context, and cited URLs.
3. Review a competitor’s cited page and your closest matching page. Mark missing facts, unsupported claims, and access problems separately.
4. Change one priority page or source at a time, note the release date, and check whether the changed URL has been recrawled.
5. Repeat the same questions over time. Compare answer presence with Search Console clicks and measured referral or lead events; do not merge those measures into one “AI rank.”

A handful of answers is a diagnostic sample, not a market-wide share of voice estimate. Responses can change with time, location, account context, and product behavior. Preserve the answer evidence so another analyst can review the result.

What should the team do next?

If a priority page is blocked or broken, fix access and internal links. If the page does not answer a buyer’s question, add the missing facts and cite their source. If the product is accurately described on-site but absent from relevant independent coverage, offer original data, a reproducible method, or a useful expert explanation to an appropriate publisher. Then measure whether the work produced qualified discovery or better answer evidence.

Digraph can help teams preserve prompt-level answer and citation evidence for this review. Start with a small, stable question set and a named competitor group. Keep the collection method and its limits beside the result.

Sources and method

This is an editorial diagnosis framework, not a controlled study of model ranking factors. The platform-specific eligibility and crawler statements above are supported by the providers’ own documentation. The suggested comparison and measurement steps are Digraph’s method for investigating answer gaps.

• Google Search Central: AI features and your websiteEligibility for supporting links in AI Overviews and AI Mode.
• Google Search Central: helpful, reliable, people-first contentGuidance on original information, clear sourcing, and content made for readers.
• OpenAI: Overview of OpenAI crawlersHow OAI-SearchBot and GPTBot are used, and how to allow or block each crawler.