Search vs Recommendation Budget Split: Where the Next Dollar Goes
Summary: Search captures intent that already exists; recommendations create intent that did not. Splitting budget between them is not a taste question — the split follows query volume, basket behaviour and where the traffic now arrives from.
Lead: most teams argue about search budgets using channel benchmarks that measure a different business. The split that works starts from two numbers: how much revenue arrives through queries, and how much arrives through browsing.
Key Takeaways
- Searchers are a minority with a majority of revenue. On-site searchers are about 24% of shoppers but drive 44% of revenue.
- Conversion gap is structural. Searchers convert about 2.5 times more than browsers, per a dataset of 609 million searches.
- AI referral is a third bucket. AI-referred traffic converted 42% better than non-AI traffic in March 2026.
- Engagement differs. AI-referred visitors browse 13% more pages and spend 48% longer per visit.
- Agent traffic converts harder. A 329-brand study measured 12.3% conversion for AI-agent visitors versus 3.1% for unassisted browsers.
- Search quality still has headroom. Better natural-language search lifts conversion 15% to 30% over keyword matching.
I. Two engines, two different jobs
Site search answers a question the visitor already formed. Recommendations answer a question the visitor has not formed yet. Both produce revenue, but they fail in different ways and are fixed with different work.
Search fails on recall and precision: the query is understood poorly, the results are sorted badly, or zero-result pages appear for phrasing that buyers actually use. The fixes are query logs, synonym coverage, and ranking signals — cheap, fast, measurable.
Recommendations fail on relevance and repetition: the same block shows the same bestsellers to everyone, so the marginal impression declines. The fixes are segmentation, freshness rules and placement testing — slower, and dependent on traffic volume.
The budget question follows from that: search investment pays back per query and can be justified with log analysis; recommendation investment pays back per session and needs enough sessions to test. A small site should fund search first, because the data exists on day one.
The third bucket is off-site but lands in the same funnel: AI referral. It behaves like a high-intent channel, and it needs a destination page that answers the question the engine quoted.
Want to know how much of the ChatGPT and Perplexity answers your brand owns?
Explore TENGENCE GEOII. An allocation model you can defend
| Stage | Site search | Recommendations | AI referral readiness |
|---|---|---|---|
| Under 50k monthly sessions | 60% | 20% | 20% |
| 50k to 250k sessions | 40% | 35% | 25% |
| Above 250k sessions | 30% | 45% | 25% |
| Query-led catalogue | 55% | 25% | 20% |
| Browse-led catalogue | 25% | 50% | 25% |
The shares are starting points, not rules. They encode one judgement: as traffic grows, recommendations earn more because there is enough session volume to test against; when the catalogue is query-led, search keeps the larger share because intent arrives already formed.
The AI referral line is not a channel budget. It funds the pages and structured facts that engines quote, so it belongs with content and technical work rather than with media spend.
Review the split quarterly against three numbers: zero-result rate, recommendation click-through by placement, and AI referral conversion by engine. A split that ignores those three is a guess with a spreadsheet.
III. What the numbers actually say
Constructor's dataset of 609 million searches across more than 100 retailers puts searchers at roughly 24% of shoppers, 44% of revenue and about 2.5 times the conversion rate of browsers. That ratio is why search upgrades usually pay back fastest: the intent is already in the building.
AI referral is the newer signal. Adobe measured AI-referred traffic converting 42% better than non-AI traffic in March 2026, with 13% more pages browsed and 48% longer visits. A separate 329-brand study measured 12.3% conversion for visitors arriving through AI agents against 3.1% for unassisted browsers.
Search technology itself still has room. Natural-language search over keyword matching has been measured at 15% to 30% higher conversion, which is a large number for a change that does not require new traffic.
Put together, these four findings support a specific order of work: fix search recall first, then add recommendation breadth, then fund the answer layer that feeds AI referral.
IV. Governance and the quarterly decision
Allocate by owner, not by tool. One person owns query quality and the zero-result rate; one owns recommendation placement and freshness; one owns the fact pages that engines quote. Three owners, one dashboard, one quarterly decision.
Set thresholds in advance. If zero-result rate stays above a target for two quarters, shift budget from recommendations to search. If AI referral conversion is above the site average and the destination pages are thin, shift budget into page work.
Avoid the annual renegotiation trap. A split fixed for twelve months is usually wrong by month five, because traffic mix moves faster than budget cycles. Quarterly review with a documented decision is cheaper than an annual argument.
Document the decision each quarter, including what was not funded. The unfunded list is the most useful artefact in the next budget cycle, because it shows the trade-offs were chosen rather than ignored.
FAQ
Q: Should search or recommendations come first?
A: Search. The query log gives you data on day one; recommendations need session volume to test properly.
Q: How much budget belongs to AI referral?
A: Treat it as content and technical work sized by the number of pages engines quote, not as a media line.
Q: Is 24% of shoppers driving 44% of revenue still true?
A: It is a large-sample finding rather than a law; measure your own split before copying the ratio.
Q: What single metric should lead the review?
A: Zero-result rate for search, placement CTR for recommendations, conversion by engine for AI referral.
Q: Can one vendor cover all three?
A: Sometimes, but the risk is that the tool defines the split instead of the data.
Q: How often should the split change?
A: Quarterly, with a written decision and an unfunded list.
About Tengence
Tengence Insight connects search logs, recommendation placements and AI referral sessions into one measurement layer, so the split is argued with the same data.
Related Reading
- SaaS content search solution
- Enterprise site search selection: Algolia and Typesense
- GEO citation monitoring dashboard
Data Sources
- Busyseed, search vs recommendations 2026: supports the 24% searcher share, the 44% revenue share, the 2.5 times conversion multiple and the 609 million search dataset.
- Value Add VC, AI product discovery 2026: supports the 42% AI-referral conversion advantage, the 13% page depth and 48% visit length figures, and the 12.3% versus 3.1% agent conversion comparison.
- RaftLabs, ecommerce automation use cases: supports the 15% to 30% conversion lift from natural-language search over keyword matching.
- SoTech, AI product discovery vs traditional search: supports the direction of the AI-referral engagement advantage used in the allocation model.