The four kinds
GEO is still a young category, and its providers come from four different backgrounds. Identify the kind before you evaluate any firm — pick the wrong kind and even thorough due diligence answers the wrong question.
1. Global SEO and content agencies with a GEO line. Strengths: mature content and digital-PR machinery, real media relationships in English-language markets. Limits: GEO is often a re-labelled SEO retainer, delivered by the old methods; measurement skews to English-language engines; facts on the Chinese side of the web are out of reach. Fits: brands whose bottleneck is English content volume and media coverage. Check: real published work in English-language media, and measurement covering the engines your buyers actually ask.
2. GEO-native boutiques (mostly overseas). Strengths: they live inside AI-answer measurement and iterate fast. Limits: small teams, no China-side capability, little ability to verify Chinese primary sources. Fits: Western brands optimising for ChatGPT and Perplexity first. Check: raw-answer samples from their measurement panel; engines matching your target markets.
3. China-based going-global specialists. Strengths: they work outward from the Chinese source facts, deliver bilingually, and cover Chinese and global engines in one measurement panel. Limits: overseas media capability varies wildly — check for real in-market operations, not resold packages. Fits: Chinese exporters whose facts, headquarters and decisions sit in China. Check: a real operating footprint in the target market (published English content, community presence), not a partner list.
4. Large consultancies. Strengths: governance, compliance, multi-market programme management. Limits: GEO becomes a slide inside a transformation deck — slow, typically senior-priced (check each firm's published rates), rarely hands-on with a crawl surface. Fits: groups buying an organisation-wide AI programme, not visibility as such. Check: does the delivery team actually change robots rules, structured data and content systems — or only advise?
Five dimensions that actually separate them
- Measurement method: a fixed question panel and engine list, with raw answers you can inspect — or screenshots and impressions? One real sampling record settles it.
- Engine and language coverage: the engines your buyers actually ask, in the languages they ask in. English-only measurement cannot answer how visible you are when Chinese buyers ask in Chinese.
- Technical remediation: can they actually change robots rules, structured data and content systems — or only advise? The weakest dimension in our 30-site measurement is machine readability (mean 69.2, median 68.5), and 17 of 30 sites had no JSON-LD at all — fixing that takes hands, not decks. Sample: 30 Chinese exporters' global website home domains, measured with the Negency AI visibility audit engine (around 30 signals, five categories) in a single-day sample on 30 October 2026; the report measures website technical visibility only, not brand mention rates in AI answers.
- Third-party mentions: directories, media, communities — the layer AI answers are actually built from. Can they show a method, not just a promise? Ask for mention samples earned for existing clients, and how each was earned.
- Transparency: fixed-price entry, published scope, clean exit terms. Who owns the data, accounts and content assets must be written into the contract.
Choose by bottleneck, not by brand
AI cannot read your site → weight technical remediation. AI describes you wrongly → weight entity-fact discipline. AI never mentions you → weight third-party distribution. HQ in China, markets abroad → weight bilingual delivery and dual-engine coverage. Most exporters' bottleneck is in the first two: in the 30-site measurement, machine readability (mean 69.2, median 68.5) and entity structure (mean 71.1) are the two weakest of the five dimensions, and 10 of 30 sites scored below 70 on both — fix the machine-readable and recognisable layers before investing in mentions.
The wrong kind has a real cost: a Chinese fact-checking bottleneck handed to an overseas boutique with no China-side capability leaves errors sitting in the citation sources; an English media bottleneck handed to a team that only rebuilds websites leaves mentions unmoved. When kind and bottleneck mismatch, the money is spent and the problem stays put.
Where Negency sits
We are deliberately in the third kind: the Chinese team verifies facts at the source; operations in Gothenburg and Mölndal work the European market in-market; one fixed panel covers Chinese and global engines in both languages; entry is a free check plus a fixed-price diagnostic (RMB 19,800, published on this site). Once you have picked a kind, run the firm through the 12 due-diligence questions — they apply to us exactly as to anyone else. Weighing software against a service? Read GEO tools vs a GEO agency first.