Back to Blog
AI SEOAEOEntity SEOCitation-Ready ContentContext MapAI Search VisibilityFractional CMO
AI search has six buyer paths. Your website sees one.
A brand can be cited, recommended, accumulated, excluded or acted upon without an attributable website visit. The mechanisms differ by vertical, and standard website analytics directly attribute only citation click-through.
AI search has decomposed the standard website funnel into at least six commercially distinct outcomes, each with different mechanics and measurement requirements. A brand can be cited without being clicked. Clicked without being recommended. Recommended without being cited. Accumulated across conversations without a single measurable referral. Excluded before comparison begins. Or acted upon inside the assistant without the brand's website involved at all. These are analytically distinct outcomes, not mutually exclusive journeys; a single buyer can pass through several of them.
Five of the mechanisms are visible in the evidence. Accumulated exposure is visible only in pieces. We can see repeated research, return visits and later branded searches. We cannot yet prove that one caused the next. Nor do we know how many businesses measure across all six.
A 2026 Semrush survey found that 45% of marketing leaders cannot accurately measure AI-answer visibility and only 9% have tools covering all relevant cross-platform metrics (Semrush, 2026). That is a measurement problem. The survey does not tell us why it exists. One plausible explanation is also the simplest: website analytics were built to see a visit, and five of these paths can happen without one.
Six buyer paths from a single AI query
Only one is reliably visible in standard analytics
AI processes query
retrieves content
Analytics visibility
1. Citation without click
cited, read, no visit
dark
registers as zero
2. Citation with click
cited, clicked, visited
visible
3. Recommendation without citation
named, no source link
dark or misattributed
shows as Direct
4. Accumulated exposure
repeated mentions, no single click
dark
attributed elsewhere
5. Exclusion
not retrieved, not compared, not presented
invisible
absence leaves no signal
6. Embedded action
no website visit required
dark to brand
platform owns the data
AI processes query
retrieves content
Path 1
Citation without click
cited, read, no visit
dark
registers as zero
Path 2
Citation with click
cited, clicked, visited
visible
Path 3
Recommendation without citation
named, no source link
dark or misattributed
shows as Direct
Path 4
Accumulated exposure
repeated mentions, no single click
dark
attributed elsewhere
Path 5
Exclusion
not retrieved, not compared, not presented
invisible
absence leaves no signal
Path 6
Embedded action
no website visit required
dark to brand
platform owns the data
This is an evidence assessment with a practical point. Claims are labeled Safe, Defensible or Reach according to how much weight the evidence can carry. The verticals are examples, not universal rules.
An AI-assisted journey can produce any of these outcomes, sometimes more than one. Standard analytics see one of them. The rest register as zero, as Direct, as someone else's traffic or as nothing at all.
1. Citation without click
A source is retrieved, cited and visible in the AI answer. The user reads the answer and does not visit the cited page. The brand contributed evidence. It received no traffic.
In Pew's Google sample, clicking a cited source was the exception, not the rule. Pew Research Center metered the browsing of 900 U.S. adults across 68,879 Google searches in March 2025 (Pew Research Center, 2025). Of those, 12,593 searches produced an AI summary. On pages where an AI summary appeared, 1% of visits led to a cited source, 8% to a conventional result and 26% ended the browsing session entirely (Pew Research Center, 2025).
That 1% is a per-search-result-page rate across all topics in a general population panel. For many searches, the citation may be the only observable source interaction. No click, no trail.
Semrush quantified the separation more precisely. In a study of 3,981 domain appearances across AI answers, 61.7% were citations without a brand mention and 25.1% were brand mentions without a citation (Semrush, 2026). Citation and brand recommendation are empirically different events. Across 1,094 ChatGPT topic clusters, only 21% of the most-cited domains were also the most-mentioned brand, and the correlation between citation leadership and mention leadership was slightly negative (-0.229) (Semrush, 2026).
A SaaS company can have its documentation cited as evidence in a comparison answer while a competitor's product is the one named and recommended. The documentation informed the answer. The brand did not benefit from it.
Where it matters most. SaaS and B2B professional services, where educational and technical content earns citations but the vendor recommendation is a separate decision. Travel and DTC retail show the same pattern when editorial or review content is cited but an OTA or aggregator receives the click.
What it requires. Content that earns retrieval for specific claims. The retrieval corpus and the conversion surface are different assets. Documentation, methodology pages, comparison evidence and technical specifications earn citations. Pricing pages, demos and booking flows convert visitors. Optimizing one as if it were both produces content that is either unreadable as editorial or unconvincing as a product page.
Status: Safe. The citation-without-click pattern is documented in metered behavioral data and large-scale content analysis. The strategic inference that this creates a measurement blind spot is Defensible.
2. Citation with click
This is the path analytics know how to see. It looks like search: a source appears, someone clicks, a session begins.
The traffic is growing. The conversion story is messier.
Adobe measured AI-referred traffic to U.S. sites across retail, travel and tech/software (Adobe Digital Insights, 2026, Adobe, 2026). The growth numbers are large: AI referrals to travel sites rose 2,215% from October 2024 to May 2026, and retail rose 1,324% over the same period (Adobe, 2026). Those are percentage increases from small, undisclosed bases. Semrush's worldwide panel of 50,000+ sites put AI referrals below 0.15% of total visits in 2025 (Semrush, 2026). The channel is growing fast. It is still small.
The conversion picture is mixed and category-dependent. During the 2026 four-day Prime Day period, Adobe found AI-referred retail shoppers converted 40% better than non-AI channels (Adobe, 2026). During the 2025 holiday period, 31% better (Adobe Digital Insights, 2026). A peer-reviewed study of 973 e-commerce sites covering August 2024 through July 2025 found the opposite direction: ChatGPT referral traffic was below 0.2% of sessions and conversion per session was above paid social but below every other traditional channel (Schulze & Kaiser, 2026). Higher product complexity improved both AI traffic share and conversion rate (Schulze & Kaiser, 2026).
The temporal contradiction matters. Adobe's more recent event-period data shows higher AI conversion. The peer-reviewed cross-country data from an earlier period shows lower. Time, event context, geography, site composition and product complexity all move the number. Neither result generalizes.
TrustRadius surveyed 2,058 verified technology buyers in January 2025 and found 90% said they clicked at least one cited source in a Google AI Overview during their purchase research (TrustRadius, 2025). Pew found 1% per search-result page (Pew Research Center, 2025). Those numbers look contradictory. They use different denominators and different methods — self-report vs. metered observation. The real conclusion is heterogeneity across populations and query types.
The cited source and click destination may be an aggregator rather than the recommended brand's website. A hotel can be recommended while the citation and click go to an OTA. A SaaS product can be named while the click goes to a review platform.
Where it matters most. Travel and hospitality, where somebody still needs live inventory, current prices, cancellation terms and loyalty treatment. That information may appear inside the assistant or require a click to a booking surface. Either way, a static answer is not enough. SaaS, where self-reported click rates among tech buyers are high and AI-referred visitors show strong engagement metrics (Adobe's tech/software panel reported 35% higher engagement, 43% lower bounce, 46% more time on site (Adobe Digital Insights, 2026)). DTC retail during high-intent shopping events.
Ahrefs reported that AI search was approximately 0.5% of its own site's visits but 12.1% of sign-ups, roughly 23 times organic's sign-up efficiency (Ahrefs, 2025). That is one company with an audience of SEO professionals who use AI tools heavily. It is a case study, not a category benchmark. A broader Amsive study of 54 validated sites found the B2B LLM conversion advantage was not statistically significant (p=.705) in the thresholded sample (Amsive, 2025).
What it requires. The retrieval corpus earns the citation. The conversion surface closes the visit. For travel: evergreen destination content earns citations, but only current inventory with total pricing and cancellation terms can close a booking. For SaaS: comparison pages and integration documentation earn citations, but a fast product page with relevant proof and a clear trial action converts. For retail: category guides and expert reviews earn citations, but the product detail page with exact variant, price, stock, delivery terms and media converts.
Status: Safe for the existence of the path and rapid growth. Defensible for conversion advantage, with material qualifications by period, category and attribution method. Reach for any universal conversion-rate claim.
3. Recommendation without citation
The AI names or recommends a brand, product or provider without citing a specific source page. The recommendation may draw on model training, retrieved web content that is not surfaced as a citation, merchant feeds, catalog data, behavioral signals, reviews or some combination of these.
Calling this "parametric recommendation from training data" is too narrow and increasingly wrong. Modern AI shopping and recommendation systems ground their answers in live data.
Amazon's Alexa for Shopping (formerly Rufus) uses catalog and store knowledge, customer activity, a product-attribute and customer-preference graph, RAG sources and multiple models (Amazon, 2026). More than 250 million customers had used it by mid-2026 (Amazon, 2026). Users were over 60% more likely to purchase during that shopping trip, though that figure reflects self-selection (people who engage an AI shopping assistant may already have purchase intent) and Amazon did not disclose adjustment methods (Amazon, 2026).
ChatGPT's shopping system uses ACP merchant feeds, public product data, retail sources, memory and clarifying questions (OpenAI, 2026, OpenAI, 2026, OpenAI, 2026). Google's AI Mode uses merchant feeds, product data, real-time pricing and personal intelligence (Google, 2026). These are grounded recommendation systems, not uncited model recall.
The measurement problem is straightforward: an unlinked recommendation may produce no referral at all, or one attributed to Direct or to a retailer. Ahrefs' 3,000-site study found that AI chatbot traffic averaged 0.17% of site visits and directly observed Copilot test clicks being recorded as Direct traffic (Ahrefs, 2025). The visible AI referral number is a floor. An aggregator can supply the comparison evidence or receive the subsequent visit even when it is not visible in the recommendation.
Where it matters most. DTC retail brands, where the AI may name the brand but send the click (and the sale) to an aggregator or marketplace. Marketplace sellers, where the recommendation happens inside the platform and the seller's own website is irrelevant. SaaS shortlists, where a product can be named in a comparison without the vendor's site being cited.
What it requires. Brand and product entities that are unambiguous, well-attested across independent sources and carry enough structured evidence to survive retrieval. For marketplace sellers: complete catalog attributes, accurate titles and descriptions, variation mapping, current availability and pricing, and review evidence. The retrieval corpus is the structured data. The conversion surface is the product detail page and offer terms. For DTC brands: presence and consistency across third-party review sites, expert coverage, creator content and comparison resources that establish the brand as a genuine option with independent attestation.
Status: Safe for the mechanism. The grounded-data correction to the original "training data only" definition is documented. Defensible for the strategic implication that brands need entity-level optimization separate from page-level citation work.
Where the value lands
A single AI recommendation can split across five different organizations
| Evidence provider | Recommended brand | Click destination | Transaction owner | Underlying data owner | |
|---|---|---|---|---|---|
Travel "best hotels in Lisbon" | Hotel blog the brand | Hotel name the brand | OTA listing aggregator | OTA aggregator | Hotel inventory owner |
SaaS "best CRM for startups" | Vendor docs the brand | Competitor someone else | Review site aggregator | Competitor someone else | Competitor someone else |
DTC Retail "best running shoes" | Review site third party | Brand name the brand | Marketplace aggregator | Marketplace aggregator | Brand product owner |
Travel
"best hotels in Lisbon"
Evidence provider
Hotel blog
the brand
Recommended brand
Hotel name
the brand
Click destination
OTA listing
aggregator
Transaction owner
OTA
aggregator
Underlying data owner
Hotel
inventory owner
SaaS
"best CRM for startups"
Evidence provider
Vendor docs
the brand
Recommended brand
Competitor
someone else
Click destination
Review site
aggregator
Transaction owner
Competitor
someone else
Underlying data owner
Competitor
someone else
DTC Retail
"best running shoes"
Evidence provider
Review site
third party
Recommended brand
Brand name
the brand
Click destination
Marketplace
aggregator
Transaction owner
Marketplace
aggregator
Underlying data owner
Brand
product owner
Your brand captures this role
Aggregator or third party
Competitor
If an aggregator holds three of those five roles, it is not just another citation target. It is part of the route to market.
4. Accumulated exposure
The user encounters a brand across multiple AI conversations, searches or sessions. No single interaction produces a click. Over time, the accumulated exposure creates enough familiarity or trust that the user searches the brand directly, visits its site through another channel, or mentions it in a later conversation where the assistant retrieves it.
This is the most intuitive path and the least proven.
No located longitudinal, user-level study demonstrates the proposed causal chain for any vertical. Current advice to correlate AI share of voice with branded search volume is a measurement hypothesis. It cannot establish that the AI mentions caused the branded searches.
The behavioral ingredients are documented. Gartner surveyed 645 B2B buyers and found they used seven sources on average during a recent purchase (Gartner, 2026). Semrush's 2026 B2B survey found that after encountering a brand in an AI answer, 71% of respondents said they visited the vendor's website, 63% searched the company on Google, 46% compared alternatives and 41% returned to the AI for follow-ups (Semrush, 2026). 6sense's global B2B buyer study of nearly 4,000 respondents found 94% used LLMs and averaged 16 interactions with the winning vendor (6sense, 2025).
These numbers support cross-channel, multi-source behavior. They do not prove that repeated AI mentions caused later direct traffic. The direct traffic could reflect brand awareness the company already had — word-of-mouth, advertising, event attendance, analyst recommendations.
Realtor.com's RealAssist documents the closest mechanical approximation: persistent memory that retains budget and priorities across sessions, with early beta users conducting 83% of interactions for context building, financial calculations, neighborhood assessment and fit evaluation (Realtor.com, 2026, Realtor.com, 2026). That is repeated interaction within a single platform's AI.
Where it matters most. Long-consideration B2B purchases (professional services, enterprise SaaS), where buying cycles span weeks or months and multiple evaluation rounds are normal. Real estate, where search often runs across dozens of sessions. Travel planning, where itinerary research can span many conversations. The common feature is a purchase cycle long enough for accumulation to plausibly occur.
What it requires. Topic-level consistency and breadth rather than single-page optimization. Semrush's ChatGPT topic study found that brands recommended in one topic cluster appeared in related clusters at a high rate (Semrush, 2026). The study cannot tell us whether depth caused the broader visibility. It does make depth the more defensible bet: build around the few topics closest to the company's actual expertise, watch what changes, then expand.
Status: Reach for the causal claim. Defensible for the measurement hypothesis. Safe for the underlying multi-source behavior.
5. Exclusion
The brand is not retrieved, not compared, not presented. The user never knows it was a candidate. Exclusion happens at multiple layers, and different layers dominate in different verticals.
Exclusion is the path that receives the least attention and may have the largest commercial consequence. Every other path assumes the brand made it into the answer. Exclusion is what happens when it does not.
The mechanisms are documented at each layer:
Crawl and index. Google requires an indexed, snippet-eligible page for AI Overview and AI Mode supporting links (Google, 2026, Google, 2025). OpenAI says OAI-SearchBot access is important for summaries and snippets (OpenAI, 2026). A blocked crawler, a noindex directive, login-gated content or non-indexable formats can remove a source before retrieval begins.
Feed and catalog. For marketplace sellers, Google Merchant Center disapproves products for feed and landing-page mismatches (Google Merchant Center, 2026). Amazon's product-search matching forms a match set before ranking it, using title, price, description and past customer actions (Amazon, 2025). ChatGPT says blocked retailer automation may cause a source to be skipped or replaced (OpenAI, 2026). Missing, stale or inaccurate structured data can prevent a product from entering the candidate set. In aggregator-dependent categories, incorrect classification, missing attributes, insufficient reviews or absence from the relevant comparison dataset can produce the same exclusion. The mechanism is credible, but its prevalence is unmeasured.
Retrieval and ranking. Eligible material must be retrieved for the original query or a fan-out subquery and survive context allocation. Ahrefs found that only 37.1% of URLs cited in AI Overviews also ranked in the conventional top ten for the same query (Ahrefs, 2026). The other 62.9% came from elsewhere. Conversely, 36.7% of cited URLs did not rank in the top 100 conventionally (Ahrefs, 2026). Retrieval for AI answers is a different selection process than traditional ranking.
Memory and preference. A remembered negative experience or stated preference can remove a brand before comparison. Amazon documents that the assistant remembers household and product preferences, including constraints like organic-only or pet-related needs (Amazon, 2026). ChatGPT's shopping documentation gives the example of omitting clown costumes if memory says the user dislikes clowns (OpenAI, 2026). Google AI Mode lets users ask it to remember or forget preferences for later use (Google, 2026).
Organizational policy. In enterprise implementations, connected company knowledge, approved-vendor lists, procurement policies, security requirements, certification mandates and prior evaluations can constrain the candidate set (OpenAI, 2026, OpenAI, 2026). OpenAI's enterprise apps are off by default, role-based access limits data sources, company knowledge is permissions-aware and memory is off by default (OpenAI, 2026). Two procurement professionals can type the same prompt and receive different eligible vendor sets because their companies impose different rules.
Regulatory and platform policy. Zillow documents a real-time Fair Housing Classifier that checks questions and responses to prevent discriminatory steering in real estate (Zillow Group, 2026). OpenAI and Google prohibit unreviewed automated high-impact housing and financial decisions (OpenAI, 2025, Google, 2026). These are necessary guardrails. They are not blanket suppression of an industry, but they can prevent specific comparison types or personalization patterns.
Where it matters most. Marketplace commerce, where match-set formation happens before the seller sees any signal. Retail and e-commerce, where feed accuracy and merchant-ranking inputs determine visibility. Travel and hospitality, where missing or stale inventory in hotel, content or price feeds can remove a property before narrative comparison. Real estate, where listing data freshness, MLS feed coverage and fair-housing classifiers shape the candidate set. SaaS and professional services, where enterprise procurement layers add an organizational exclusion gate invisible to public prompt tracking.
What it requires. Diagnosis before optimization. The first question is whether the brand can be found at all, not whether it ranks well. For web-based exclusion: crawler access, indexability, snippet eligibility and structured data accuracy. For marketplace and commerce: feed completeness, attribute matching, price and availability accuracy, and review and behavioral evidence (new products face a cold-start disadvantage from missing behavioral history, with Amazon noting temporary adjustments sometimes used (Amazon, 2025)). For enterprise: machine-verifiable compliance evidence, because a certification buried in a PDF that names the parent company but not the operating subsidiary may not survive automated procurement matching.
Status: Safe for the mechanisms. Defensible for the strategic importance. No source quantifies the prevalence of exclusion at any layer across any vertical.
6. Embedded action
Embedded action is broader than checkout. The assistant may complete a purchase, begin a booking, schedule a tour, connect the user with an agent or open a merchant-controlled checkout experience. The common feature is that a conventional visit to the brand's website is no longer required.
Amazon's Alexa for Shopping can add to cart and auto-buy within the assistant experience (Amazon, 2026). Google AI Mode can compare real-time flight and hotel data, track flight prices, show points and miles rates and support hotel booking, drawing from more than 300 partner airlines and travel sites (Google, 2026). Zillow's AI mode uses live for-sale and rental listings and can compare homes, explore affordability, book tours and connect users to agents (Zillow Group, 2026). Realtor.com's RealAssist retains budget and priorities across sessions and connects users to agents (Realtor.com, 2026, Realtor.com, 2026). OpenAI's March 2026 product-discovery update prioritizes merchant-owned checkout and in-app browser experiences (OpenAI, 2026).
Sometimes the transaction occurs on the platform. Sometimes the assistant hands the user into an embedded or merchant-controlled experience. Either way, the old website session is no longer the necessary bridge between recommendation and action.
Where it matters most. Marketplace commerce, where the transaction has always occurred on the platform and AI simply adds a conversational interface. Real estate, where portal AI can now complete tour scheduling and agent connection that previously required navigating the portal's conventional interface. Travel, where flight and hotel booking inside the AI answer is a documented, expanding capability. DTC retail, where agentic checkout through merchant-owned experiences is emerging.
What it requires. For marketplace sellers: the same catalog and feed optimization as the recommendation and exclusion paths, plus offer competitiveness at the moment of comparison (price, availability, fulfillment, return terms). For travel: accurate, current inventory in hotel, content and price feeds, plus integration with booking and availability APIs. For real estate: live listing data in portal and MLS feeds, accurate property attributes and current status. Here, the retrieval corpus and conversion surface begin to converge. The same price, availability, attributes and offer terms that earn consideration can determine whether the action is presented.
Status: Safe for the mechanism and its growing importance. Product documentation from Amazon, Google, Zillow, Realtor.com and OpenAI all describe these capabilities (Amazon, 2026, OpenAI, 2026, Google, 2026, Zillow Group, 2026, Realtor.com, 2026). Independent adoption and conversion benchmarks are weak: Amazon's 60% purchase-likelihood association is self-reported with undisclosed adjustment methods (Amazon, 2026), and no property-level or hotel-level incrementality study was located.
Which of these paths actually carries commercial weight for your business? Plate Lunch Collective's six-path diagnosticmaps where a brand appears, where intermediaries capture the value and where exclusion occurs before comparison.
Where the evidence breaks
Three cross-cutting problems affect all six paths.
Growth from small bases. Triple- and quadruple-digit percentage growth in AI referrals starts from tiny absolute numbers. Semrush put 2025 worldwide AI referrals below 0.15% of visits (Semrush, 2026). The peer-reviewed e-commerce study put ChatGPT below 0.2% of sessions through August 2025 (Schulze & Kaiser, 2026). Growth rate and channel share are different measures. Growth rate is the one that gets quoted.
Conversion selection bias. People who click after reading an AI comparison may already want to buy. Adobe's high-conversion holiday and Prime Day results may reflect deal-seeking behavior, returning-customer composition or large-retailer site quality — not an inherent property of AI referrals (Adobe Digital Insights, 2026, Adobe, 2026). The peer-reviewed study explicitly warns that last-click attribution understates discovery effects (Schulze & Kaiser, 2026).
Platform drift. Models, source pools and interfaces change fast. Semrush observed large within-weeks citation-source shifts (Semrush, 2025). Ahrefs found that AI Overview and organic overlap fell from approximately 76% in July 2025 to 38% in March 2026 (Ahrefs, 2026). Any static claim about what earns retrieval has a shelf life measured in weeks.
A fourth problem is vertical resolution. The published panels do not isolate the verticals practitioners care about. Adobe's retail sample includes large retailers alongside DTC brands (Adobe Digital Insights, 2026). The peer-reviewed study mixes retailers, marketplaces and lifestyle sites (Schulze & Kaiser, 2026). "SaaS" is rarely a panel category. "Professional services" has the weakest evidence of any vertical — no located services-only AI traffic, citation or conversion benchmark. Real estate outcome data is almost entirely portal press releases (Zillow Group, 2026, Realtor.com, 2026). The numbers exist. The vertical breakdowns do not.
Evidence status by path
Path
Mechanism
Commercial impact
Causal attribution
Best data source
Citation without click
Safe
Defensible
N/A (no click to attribute)
Citation with click
Safe
Defensible(category-dependent)
Last-click only
Recommendation
Safe
Defensible
Weak (referral misattribution)
Accumulated exposure
Safe(behavior)
Reach (causal claim)
None located
Exclusion
Safe(mechanisms)
Defensible(importance)
N/A (absence is the outcome)
Embedded action
Safe
Reach (conversion)
None independent
The table compresses the status labels. Safe: documented and reproducible. Defensible: strong inference from documented evidence. Reach: plausible but unproven. No path has Safe evidence for causal commercial attribution. The strongest quantitative data belongs to DTC/retail (the peer-reviewed 973-site, 10.5-billion-session study (Schulze & Kaiser, 2026)) and citation with click (Adobe's longitudinal panels (Adobe Digital Insights, 2026, Adobe, 2026, Adobe, 2026)). The weakest belongs to accumulated exposure and embedded-action conversion.
Working priorities by vertical
The mix varies by business type. The table maps where the evidence points, where the leverage sits and where the gaps are largest. These are investigation priorities, not measured shares of buyer journeys.
Citation without click
Citation with click
Recommendation
Accumulated exposure
Exclusion
Embedded action
SaaS
Core
Core
Supporting (shortlists)
Unmeasured
Supporting (enterprise procurement)
Unmeasured
E-commerce marketplace
Supporting
Supporting
Core
Unmeasured
Core (match-set/feed)
Core
B2B professional services
Supporting
Supporting (validation)
Unmeasured
Unmeasured
Supporting (organizational policy)
Unmeasured
Travel and hospitality
Supporting
Core (live terms)
Supporting
Unmeasured
Core (feed/inventory)
Core and growing
DTC and retail
Supporting
Core (category-dependent)
Core
Unmeasured
Core (feed/merchant)
Emerging
Real estate
Unmeasured
Unmeasured
Supporting (portal AI)
Emerging
Core (data/policy)
Core for portals
What the table compresses. Complex products benefit more from AI referrals in DTC retail, in both traffic share and conversion (Schulze & Kaiser, 2026). Enterprise procurement exclusion in SaaS is underexamined but credible (OpenAI, 2026, OpenAI, 2026). Professional services has the weakest evidence base of the six verticals. Travel's embedded-action path moved from documented to operational in August 2026 when Google AI Mode added flight and hotel booking from over 300 partner sources (Google, 2026). Real estate outcome data is almost entirely portal press releases (Zillow Group, 2026, Realtor.com, 2026).
What remains unknown
The evidence report behind this article identified specific unknowns for each vertical. The ones that matter:
No study connects AI citations to paid SaaS revenue or retention. No independent seller-level study reports AI recommendation, exclusion or conversion rates in any marketplace. No professional-services-specific AI traffic or conversion benchmark exists. No hotel-level or property-level study links AI citations to incremental bookings. No clean DTC-only panel isolates DTC conversion from large-retailer data. No independent AI-generated lead or closing-rate benchmark exists for real estate. And no study in any vertical randomizes brand inclusion and exclusion in AI answers and measures later purchase. Almost all commercial-effect evidence is correlational.
The multi-touch causal chain — repeated AI mentions cause later branded search — remains unproven in every vertical. It is the most commonly asserted and least evidenced claim in AI search strategy.
The diagnostic question
Seven questions. Each one produces different work.
- When we are cited, does our brand get named? (Citation without click vs. ghost citation.)
- When we are cited and clicked, what happens on the page the user reaches? (Citation with click, and the gap between retrieval corpus and conversion surface.)
- When we are recommended without citation, where does the traffic land? (Recommendation, and whether an aggregator or marketplace captures the conversion.)
- Are we present across the topic clusters where our expertise applies? (Accumulated exposure, and topic-level consistency.)
- Can we be found at all? (Exclusion, at every layer from crawl to enterprise policy.)
- When the AI completes the action, are we in the candidate set with competitive terms? (Embedded action, and the convergence of retrieval and conversion data.)
- Who controls the evidence and captures the value? (Record the cited source, source owner, recommended entity, click destination and transaction owner separately. They may be five different organizations.)
The first move for each:
- Cited but not named. Audit your entity presence across the independent sources the AI retrieves from. If your brand appears only in your own documentation, the citation earns evidence credit but the recommendation goes to whoever has third-party attestation.
- Cited and clicked, but the page doesn't convert. Map the gap between the pages that earn citations and the pages that close. If the AI sends someone to a methodology article and the next step is unclear, the click was wasted. The retrieval corpus and the conversion surface are different assets. They need a visible bridge between them.
- Recommended but traffic lands elsewhere. Track where the visit goes. If the AI names your brand but the click routes to an aggregator, marketplace or retailer, the recommendation is working for someone else. The fix is entity-level optimization and direct-channel strength.
- Present in one cluster but absent from adjacent ones. Start with the few clusters closest to the highest-value conversion. Build depth, watch what changes, then expand.
- Not found at all. Start at the bottom of the stack: crawler access, indexability, snippet eligibility, structured data accuracy. For marketplace sellers, start with feed completeness and attribute matching. For enterprise sales, check whether your compliance evidence is machine-verifiable or buried in a PDF the procurement system cannot parse.
- Not in the candidate set when the AI completes the action. The conversion surface is the structured data itself: catalog attributes, offer terms, availability, pricing. If that data is incomplete, stale or uncompetitive at the moment of comparison, no amount of content optimization upstream will matter.
- The value lands somewhere else. Map five roles for every commercially important answer: evidence provider, recommended brand, click destination, transaction owner, underlying data owner. If an aggregator holds three of those five roles, it is not just another citation target. It is part of the route to market.
Measuring one path while the business outcome depends on another produces activity that looks productive and changes nothing.
Sources
Every source below is cited inline throughout the article. Status labels (Safe, Defensible, Reach) are assigned in the body text above.
- Pew Research Center. "Google users are less likely to click on links when an AI summary appears in the results." 22 Jul 2025. pewresearch.org
- Semrush. "Why 62% of AI citations don't lead to brand mentions." 9 Jun 2026. semrush.com
- Semrush. "AI visibility is a topic-level game: A study of 50,000 brands in ChatGPT." 20 Jul 2026. semrush.com
- Semrush. "Semrush Releases Expanded 2026 AI Visibility Index." 2026. semrush.com
- Semrush. "We analyzed billions of web visits: How AI is reshaping traffic channels." 27 Apr 2026. semrush.com
- Semrush. "How AI tools shape the B2B buying process." 8 Jul 2026. semrush.com
- Semrush. "Semrush AI Overviews Study: What 2025 SEO Data Tells Us About Google's Search Shift." 15 Dec 2025. semrush.com
- Semrush. "AI Overviews are expanding across commercial intent search." 2 Jul 2026. semrush.com
- Adobe Digital Insights. "AI-sourced traffic insights 2025." 4 Feb 2026. adobe.com
- Adobe. "Introducing Adobe Brand Visibility." 17 Jun 2026. adobe.com
- Adobe. "AI shoppers convert 40% better as Prime Day hits $26.4B." 29 Jun 2026. adobe.com
- Christian Schulze and Maximilian Kaiser. "ChatGPT Referrals to E-Commerce Websites." Marketing Science, 21 Apr 2026. pubsonline.informs.org
- Amsive. "Does LLM Traffic Convert Better Than Organic?" 3 Sep 2025. amsive.com
- TrustRadius. "Bridging the Trust Gap: B2B Tech Buying in the Age of AI." 7 Apr 2025. trustradius.com
- Gartner. "69% of B2B Buyers Turn to Sales Reps to Validate AI-Generated Insights." 20 May 2026. gartner.com
- 6sense. "The B2B Buyer Experience Report for 2025." 2025. 6sense.com
- Ahrefs. "63% of Websites Receive AI Traffic." 6 Feb 2025. ahrefs.com
- Ahrefs. "Update: 38% of AI Overview Citations Pull From The Top 10." 2 Mar 2026. ahrefs.com
- Ahrefs. "The ChatGPT Traffic Playbook." 2025. ahrefs.com
- Amazon. "Amazon's next-gen AI assistant for shopping." 13 May 2026. aboutamazon.com
- Amazon. "Amazon compliance report 2025, Annex 1." 2025. aboutamazon.com
- OpenAI. "Using shopping research in ChatGPT." Updated Aug 2026. help.openai.com
- OpenAI. "Shopping with ChatGPT Search." 2026. help.openai.com
- OpenAI. "Powering Product Discovery in ChatGPT." 24 Mar 2026. openai.com
- OpenAI. "Publishers and Developers FAQ." Updated Aug 2026. help.openai.com
- OpenAI. "Company knowledge in ChatGPT." Updated Aug 2026. help.openai.com
- OpenAI. "Admin Controls, Security, and Compliance." Updated Aug 2026. help.openai.com
- Google. "Optimizing your website for generative AI features." Updated 2026. developers.google.com
- Google. "AI features and your website." Updated 10 Dec 2025. developers.google.com
- Google. "3 new ways to plan and book travel in Search." 27 Aug 2026. blog.google
- Google. "Personal Intelligence in AI Mode." 22 Jan 2026. blog.google
- Google Merchant Center. "Tips to help your products stay approved." 2026. support.google.com
- Google Hotels. "Hotel APIs." 2026. developers.google.com
- Zillow Group. "Zillow debuts AI mode." 25 Mar 2026. investors.zillowgroup.com
- Realtor.com. "Realtor.com Launches RealAssist AI." 2 Jun 2026. mediaroom.realtor.com
- Realtor.com. "Quarterly product update, July 2026." realtor.com
- Realtor.com. "82% of Americans Use AI for Housing Market Information." 9 Oct 2025. mediaroom.realtor.com
- OpenAI. "Usage policies." Effective 29 Oct 2025. openai.com
- Google. "Generative AI Prohibited Use Policy." 2026. policies.google.com
Ready to appear in AI search?
We work with businesses across every industry. If you have questions about where you stand in modern search, we are easy to reach.
Hayden Bond
Hayden Bond has been doing SEO since 2004. He founded Plate Lunch Collective in Aiea, helping brands get cited by AI platforms rather than just ranked by Google.
