Where the Value Lies in a Multi-Agent Pipeline for Generative Engine Optimization: Evidence from Five Products and a Field Deployment

Answer engines such as Doubao and Gemini now answer many product questions directly, so organisations need content these engines will cite. Producing it is a business process: choosing questions, writing, checking, adapting to many platforms, publishing and measuring. We built GEO-Loop, a multi-agent pipeline that runs this process with 23 agent roles around deterministic state, an automated release gate and evidence-based publication records, and asked where its value lies. On 20 topics for five products in Chinese and English, a single prompt that knew the gate's rules passed 17 times, as often as the pipeline, but by leaving out sourced and dated facts; asked to revise itself, it fell to 7, mostly by inventing numbers. The pipeline's first drafts passed 8 times; repairing only the failed checks raised this to 17 while keeping their evidence, and 19 of 20 platform versions passed after one repair. In a 19-day deployment the pipeline produced 164 pieces on ten channels for about US$7, of which 43% could be verified as published. Its value lies in the repair loop and in the stages a prompt cannot do, not in better first drafts, and pass rates alone favour articles that omit evidence.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-18
DOI
https://doi.org/10.5281/zenodo.24352477
Primary Topic
Information Retrieval and Search Behavior
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Where the Value Lies in a Multi-Agent Pipeline for Generative Engine Optimization: Evidence from Five Products and a Field Deployment

Lintong Han, Mengran Wang
3 links
Zenodo (CERN European Organization for Nuclear Research)
Information Retrieval and Search Behavior
preprint

Where the Value Lies in a Multi-Agent Pipeline for Generative Engine Optimization: Evidence from Five Products and a Field Deployment

Lintong Han, Mengran Wang
preprint en

Abstract

Answer engines such as Doubao and Gemini now answer many product questions directly, so organisations need content these engines will cite. Producing it is a business process: choosing questions, writing, checking, adapting to many platforms, publishing and measuring. We built GEO-Loop, a multi-agent pipeline that runs this process with 23 agent roles around deterministic state, an automated release gate and evidence-based publication records, and asked where its value lies. On 20 topics for five products in Chinese and English, a single prompt that knew the gate's rules passed 17 times, as often as the pipeline, but by leaving out sourced and dated facts; asked to revise itself, it fell to 7, mostly by inventing numbers. The pipeline's first drafts passed 8 times; repairing only the failed checks raised this to 17 while keeping their evidence, and 19 of 20 platform versions passed after one repair. In a 19-day deployment the pipeline produced 164 pieces on ten channels for about US$7, of which 43% could be verified as published. Its value lies in the repair loop and in the stages a prompt cannot do, not in better first drafts, and pass rates alone favour articles that omit evidence.

Zenodo (CERN European Organization for Nuclear Research)
Information Retrieval and Search Behavior
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon