Brand growth in the AI era is entering a new phase: being recommended by AI

01 Facts & context

Platform guidance: discoverable, verifiable brand facts

Google’s official guidance says AI Overviews and AI Mode use foundational SEO requirements, without special AI files or markup; meeting requirements does not guarantee inclusion. This supports concentrating on reliable information rather than promising that a particular technique will consistently produce recommendations.

From a brand-management perspective, the website should be a public source that people and external systems can trace back to. Important information needs clear pages, headings and internal links. Products and services should be explained in readable content, supporting evidence should be accessible, and someone should maintain outdated material. Visitors need help deciding whether the offer suits them, not a wall of jargon.

A service page can explain relevant problems, deliverables, collaboration and the next step. A product page can explain specifications, conditions of use and after-sales support. Structured information should describe this visible substance rather than invent promises absent from it. The website’s role then extends beyond image-building to a continuously maintained source of credible explanation.

02 GSP analysis

An extra layer of explanation before the choice

Questions put to AI often combine goals, constraints and trade-offs: suitable for a small space, easy to maintain, within a budget, or different from several alternatives. A brand therefore faces more than an opportunity to appear for a keyword. It participates in a process of explanation, screening and comparison. Entering the shortlist, and the reason for entering it, become management questions.

We use “recommended by AI” to mean an accurate, relevant and evidenced explanation or reason for choosing in a situation consistent with the brand’s actual capabilities. A mention, a citation, a recommendation and a purchase are different outcomes. A brand may be mentioned only as background, while a citation may simply support an isolated fact.

Traditional marketing still matters. Product experience, brand memory, channel service and real customer word of mouth affect the final decision. The additional task is to make these existing sources of value easier to understand and verify, then connect interest generated by an answer with actual products, service and relationships.

03 GSP analysis

Turn delegated decision-making into tasks the organization can deliver

In his Zhen Brand Talk presentation, GSP founder Deng Fei proposed that consumers increasingly delegate parts of their judgment to AI. He discussed five brand capabilities: expertise, reliability, insight, ongoing support and transparency. The presentation took place on September 10, 2026, and the original WeChat article was published on September 14. We use this as a GSP framework for investigating brand capability, not as a statistical conclusion about every consumer or as a scoring formula disclosed by an AI model. The distinction matters when a proposition becomes an implementation plan.

Brands can decide whether their information is clear and verifiable, but they cannot control how a platform selects sources in every response. Information and service quality are manageable inputs; recommendation outcomes also depend on the question, context, product coverage and platform behavior. A plan should therefore specify which inputs can improve, which outcomes must be observed and which constraints are external. It should not promise that raising an invented trust score will guarantee recommendation. Separating controllable work from external outcomes gives teams a defensible responsibility and prevents an attractive framework from becoming an unsupported performance commitment.

Expertise can be expressed through explaining difficult choices. An equipment supplier might describe operating environments, important specifications, maintenance needs and unsuitable uses, with evidence behind material claims. Reliability requires the same model's specifications, commercial conditions and service coverage to agree across pages, with a named owner for updates. Insight concerns what a real question means. A procurement buyer asking for a cheaper option may be concerned about energy, downtime or training rather than the quoted price alone. Establishing that meaning determines which information is missing and which page could actually help the buyer choose.

Ongoing support and transparency extend the work beyond enquiry. If content helps someone select a solution, service must explain implementation conditions, resolve problems and maintain useful documentation afterward. A recommendation based on conditions that cannot be delivered may create unsuitable leads instead of valuable demand. Transparency does not require publishing all internal information. It requires enough accurate detail about capabilities, conditions and boundaries for a reasonable choice. Customer material should be presented only within an authorized and clearly described case; invented client identities or performance results do not become acceptable because they make the page sound persuasive.

Under this approach, brand, product, operations, service and technology jointly maintain a fact register. Each important statement has an owner, evidence location, update date and applicable scope. Editors then organize those statements around understandable customer questions. New material should address consequential gaps in a decision before pursuing publication volume. This working arrangement can improve information supplied to people and machines, while actual enquiries and subsequent relationships determine its business value. Neither the framework's name nor one search screenshot can establish that value, and the record should remain open to correction when customer behavior contradicts the team's initial explanation.

04 GSP analysis

Establish consistent brand facts first

Many brands have abundant content but inconsistent facts. A service description changes on the website while old promotional pages remain. Channels describe the same specification differently. A case study states an outcome without explaining the process, conditions or source. Such inconsistencies increase the work required by both people and systems to interpret information.

Begin with a fact inventory: who the business is, whom it serves, the problems it solves, what the offer includes, conditions affecting suitability, evidence available for verification and ownership of updates. Important information needs a corresponding public explanation, rather than existing only in internal presentations.

Distinguish current capabilities, future plans and research perspectives. Explain the actual scope of involvement in client work and identify the period and definition behind cited data. This basic work gives comparison a reliable foundation and helps sales, service and communications use the same explanation.

05 GSP analysis

Google and Bing measure different things: citations are not sales

Before comparing AI visibility across platforms, establish what each tool records. Google Search Central describes AI Overviews and AI Mode search activity within Search Console's Web reporting. Bing's 2026 AI Performance preview reports citation activity across supported AI experiences and explicitly distinguishes citation counts from rank, placement or authority in an individual answer. The products have different populations and coverage. Adding their numbers does not create a meaningful measure of recommendation across the whole web, and a changing product definition can also alter what a time-series comparison means.

Separate discovery, citation, recommendation and business outcomes. Crawling indicates a possibility of retrieval. Inclusion in a citation list means that a page served as a reference in a response. Explicitly presenting the brand as suitable for a particular need is closer to recommendation. A visit, enquiry or purchase is a subsequent behavior. There may be losses between these stages, and some useful encounters generate no traceable click. Growth at one stage cannot simply be projected into the next; citation counts are especially unsuitable as a substitute for acquired customers or for incremental revenue attributed to content.

Questions used for monitoring should reflect actual decisions. Asking a system to recommend a named brand introduces the brand into the prompt and says little about unprompted discovery. Separate open needs, comparisons between alternatives, brand verification and questions about use. Record relevant language, geography and access conditions along with the question version and observation time. Then examine whether answers accurately express the offer's scope, omit necessary conditions or use outdated material. An aggregate ratio without this context cannot identify which content should change or distinguish a newly discovered audience from a change in the measurement procedure.

Platform reports and manually sampled questions complement one another but do not represent the same population. Reports reflect activity defined by a platform; samples reveal wording and how claims appear in individual answers. With a small sample, absence shows only what happened under those particular conditions. Changing questions, regions or reporting periods should be disclosed in comparisons. A brand named alongside an incorrect specification is an information problem. A page used to explain the category without naming the brand may still have contributed to understanding. Collapsing both situations into a binary success measure discards the evidence needed to improve the page.

On the business side, examine whether enquiries fit the offer. Record source information, needs and later outcomes within existing analytics and customer conversations, accepting that some journeys cannot be fully attributed. If a frequently cited page generates questions about services the firm does not provide, clarify its boundaries before celebrating visibility. A page with fewer visits but better-informed enquiries may warrant further investigation. The aim is to improve assistance with a decision, using several connected observations, rather than to find a single flattering measure that replaces all the outcomes the business ultimately needs to understand.

06 GSP analysis

Organize content around decisions

Someone comparing alternatives needs more than a general discussion of industry trends. Move from what the company wants to say to what the person needs to resolve: how to choose, who an offer suits, what differs, what implementation requires and when another option should be considered. Specific questions create room for distinctive, useful information.

Google’s guidance on helpful content emphasizes people’s needs, original information and substantive value. For a brand, the implication is to add understanding rather than repeat existing conclusions. Expert product explanations, verifiable accounts of project work and clarification of common misunderstandings are assets worth developing. Repetition under new headlines adds far less.

Content production also needs feedback from the business. Turn recurring sales questions, service explanations and genuine customer concerns into maintained knowledge. After publication, examine whether it reduces communication effort, improves understanding and attracts better-matched enquiries. Article volume measures workload; helping people make decisions measures value.

07 GSP analysis

After machines understand the offer, can the business deliver it?

On product and service pages, content work quickly becomes a question of who guarantees the information being communicated. Google's product structured-data guidance describes details such as price, availability, delivery and returns, with different markup appropriate to different page types. It also discusses the complementary use of page data and Merchant Center feeds. These are requirements and presentation opportunities within a particular search ecosystem. They are neither a universal interface for every AI platform nor a promise that implementing markup will secure a recommendation. The underlying operating questions remain even if the visible search format changes.

Those fields expose a common ownership problem. Content teams often do not control current product facts. Inventory belongs to transaction systems, delivery coverage to operations and service conditions to business owners, while website descriptions may be maintained manually by different people. If a person or an AI system reads obsolete information, the resulting enquiry can conflict with actual fulfillment. Begin by identifying the authoritative origin and update responsibility for each fact, then decide how to provide it through visible copy, data fields or interfaces. Correct formatting without a dependable source can simply distribute an error more efficiently.

Service businesses face a similar handover despite having no inventory field. A consulting page should explain the problem addressed, work included, required inputs and form of delivery instead of listing broad capability words. If pages promise different timeframes or deliverables for the same engagement, the customer must reconstruct the offer during the sales conversation. Maintain a business-approved service description covering common misunderstandings, optional scope and unsuitable conditions. Where pricing depends on circumstances, explain the factors and enquiry process rather than inventing a universal figure to fill a field that does not fit the service model.

Case content also needs to be checkable. A useful study describes its subject, context, actions, sources and analytical boundaries. A public company example can support learning without implying that GSP worked on the engagement. A proposed scenario can explain a method while making clear which steps are recommendations. Where actual results are unavailable, show observations that would be useful and alternative explanations that should be investigated; do not convert expected benefits into realized outcomes. Clear separation of reported facts and analysis helps customers judge fit and reduces the chance that a machine will restate a proposed action as a historical event.

Finally, create a route for correction. If service teams repeatedly encounter a misunderstanding caused by a page, they should reach the person responsible for the underlying fact. When a product condition changes, inspect related questions, case references and data fields together. After publication, continue observing whether the offer is described accurately and which information prospective customers still lack. AI visibility work then connects with product and service management rather than becoming a detached publishing exercise. Reliable information and delivery capability can accumulate over time even while platforms change their presentation; customers still need someone to fulfill the expectation created when the offer was understood or recommended.

08 Suggested actions

Measure the quality of recommendations, not only appearances

Appearing in one answer does not show that a brand will be recommended across questions, models and dates. A single absence does not prove that every content improvement failed. AI answers vary, so monitoring needs recorded conditions and a repeatable method.

Build a question set covering awareness, comparison and selection. Record the wording, platform, time, response and cited sources. Evaluate separately whether the brand is mentioned, accurately described, appropriately shortlisted and supported by credible reasons, and whether citations lead to valid information. Classifying errors helps distinguish missing facts, difficult-to-find content and situations in which the brand simply does not fit.

At the business level, assess whether visitors find what they need, whether enquiries match the service scope and whether sales and service can handle the next step. Do not equate a position in a model’s answer directly with revenue contribution. Connecting answer quality, customer behavior and subsequent business outcomes reveals which part of the process an investment has improved.

09 Suggested actions

Connect the new entry point to lasting growth

Start with a defined business problem rather than immediately expanding content production. Select a core product or service, establish important questions and verify public facts. Improve the pages, evidence and service information that most affect understanding. Then repeat the same evaluation and observe changes in real enquiries. Each step needs ownership and records that can be reviewed.

Brand, product, content, technology and customer-service teams all have a role. Marketing can organize questions and expression, operating teams confirm capabilities and delivery, technical teams keep information accessible, and service teams return real feedback. Together they help the brand be described accurately in external answers and fulfill its promise when people make contact.

An AI recommendation is a new entry point for growth, while repeated choice still depends on lasting value. Aligning the recommendation with real experience, and maintaining a feedback loop between content and operations, gives visibility a chance to become trust and trust a chance to support higher-quality loyalty growth.

Accurate understanding supports consideration; consistently delivered value supports repeated choice.