AI search: the business behind the answer
How advertising, referrals and subscriptions compete for value when discovery becomes conversational.
Public-source desk research, checked 16 October 2026. Company disclosures describe vendor claims and published terms. Commercial conclusions are Offaxis interpretation. Numerical scenarios are explicitly synthetic. We have not independently tested products or audited private company financials.
The commercial question
Our thesis: an AI answer becomes commercially valuable when it changes a purchase decision that can be measured. Conversation length, answer quality and advertising yield are different variables. A service can improve the first two without proving the third. This report examines the business architecture, rather than estimating undisclosed revenue or search market share.
Evidence: a platform is testing the transaction path
Google’s May 2026 advertising announcement describes tests of conversational ad formats and an expansion of its Direct Offers pilot. It also describes native checkout integration for eligible merchants. This is evidence of product direction and announced availability, not proof that every advertiser can access every format or that the formats improve profit. [1]
Its February 2026 commerce letter describes sponsored retailer placements and tests beyond retail. Its May 2025 announcement had already described ads within or alongside AI search responses. Together these establish a sequence of commercialization efforts; they do not isolate AI search revenue from the wider advertising business. [2] [3]
Four models, four economic constraints
| Model | Who pays? | What must work? | Risk |
|---|---|---|---|
| Advertising | Advertiser | Measurable incremental demand | Revenue rewards placements that weaken trust |
| Referral or transaction | Merchant or marketplace | Attribution and fulfilled purchases | Returns, cancellations and channel dependence |
| Consumer subscription | Reader or researcher | Repeated value sufficient for renewal | Free substitutes and expensive intensive users |
| Specialist business search | Organization | Useful coverage, reliable retrieval and workflow fit | Data costs and narrow demand |
These are commercial possibilities, not a claim that all AI search services currently offer all four.
A merchant should measure contribution
Imagine an illustrative campaign with 1,000 paid visits at $1 each, a 3% completed-order rate and $50 contribution per order before acquisition expense. Thirty orders produce $1,500 contribution; after $1,000 advertising expense, $500 remains before other campaign costs. At a 2% order rate the same arithmetic reaches zero. The example is synthetic and does not predict any platform’s performance.
The analyst must then ask whether those orders were incremental. If the campaign merely intercepts customers who would otherwise have bought directly, attributed sales overstate its value. Returns, discounts, fulfilment and repeat purchases also change the calculation. A lower click count could be acceptable if the remaining visits convert better; a higher count could be unprofitable.
The competitive position
Our interpretation is that an incumbent’s advantage can come from advertiser relationships, merchant information and distribution as much as from answer generation. A new entrant needs a reason for users to return and a route to commercial inventory or direct payment. A specialist service may find that reason in a difficult, recurring research problem. Its market may be smaller, but willingness to pay can be tested against a specific job.
Publishers occupy a different position. They supply information and depend on discovery, yet may receive fewer visits when an answer resolves the question. Offaxis would evaluate a publisher by direct subscriber acquisition and distinctive source material alongside referral traffic. Being mentioned in answers is not, by itself, a revenue model.
The trust constraint and the next evidence
Sponsored material needs a clear distinction from the editorial answer. Google says its new formats remain labeled as sponsored; whether users understand that distinction is a separate empirical question. [1] Our commercial hypothesis is that unclear incentives can reduce repeat use, making short-term monetization an incomplete measure.
We would monitor merchant access, disclosed rollout geography, incremental conversion experiments, acquisition contribution after returns, subscriber retention and the proportion of use that requires expensive research. A stronger evidence base would report comparable cohorts over time. Until then, product announcements support a commercialization thesis, not a reliable estimate of profitability.
Sources and evidence
Stay in the loop
Sign up for Offaxis news and updates. Free email signup; paid membership is separate.