InstinctUNOFFICIALAn Any Inkling project
← Back to News & analysis
22 October 2026

An AI assistant's results depend on how businesses handle discovery, requests and permission

The shift happens before checkout

The shopping shortlist is moving out of the search box.

In Capgemini's survey, the share saying Gen AI replaced search for product recommendations more than doubled.

202325%
→
202458%

Replaced search with Gen AI for recommendations · % · scale 0–100

2023
25%
2024
58%

+33 percentage points. Not a forecast. A reported change in how people look for products.

Why it matters

Getting onto the shortlist comes before getting the sale.

Our reading: use an assistant to explore options, then choose what fits your needs.

Capgemini Research Institute · 9 Jan 2025 release. Latest survey: 12,000 adults in 12 countries, Oct-Nov 2024. Self-reported recommendation search, not completed purchases or proof of recommendation quality.
Read the survey and methodology ↗

Sources: Business Insider ·21Oct2026 · Opus Research ·21Oct2026 · Decagon PACT proposal ·6Oct2026

A personal agent can make a request, but a business still controls what gets found and what actions it accepts. Reports on shopping discovery, customer-service demand and delegated authority examine different parts of that handoff. This editorial connects them; it does not claim the sources jointly measured one outcome.

Our reading

The next test is the handoff: which products get found, which requests get handled, and what a business will let an agent do.

Before the purchase

Before the purchase: Business Insider describes Azoma's work helping brands become discoverable to shopping tools. That is a reason to ask how a recommendation fits your brief, not proof that any particular assistant's results are manipulated. Instinct is a passing market example, not audited in the report.

After the purchase

After the purchase: Opus points to billing disputes, warranties and repeat contacts. Answering a question is different from completing a requested action. More persistence could bring more service work, but the article presents expectations, not a measured increase or proof a specific request succeeds.

At the handoff

At the handoff: on6October, Decagon described PACT, co-developed with Instinct, separating an agent's identity from what its customer authorized. This is a dated vendor proposal, not evidence every business supports it or today's task uses it.

What it means

Our reading: assistant quality cannot be judged from chat alone. The business side influences discovery, handling and permission. This is our interpretation across sources, not a finding they jointly measured.

Try this

Try one bounded task. Name must-haves and approval limits, then check the result at the original business. Ask what completed, what remains and what evidence supports it.

Business Insider ·21Oct2026

Opus Research ·21Oct2026

Decagon PACT proposal ·6Oct2026

Original analysis. Conceptual synthesis, not measured outcomes or a guarantee of adoption.