First, a change in how buyers buy: they ask AI

Before procurement and marketing teams build a vendor longlist, more and more of them now ask ChatGPT, Perplexity or Doubao first: "Which companies do GEO?" "Which consultancy actually understands Chinese companies expanding overseas?" The AI names only a handful of providers, and its answers come from public information across the web — not from ad placements. We ran a baseline sample of 30 fixed questions on two leading engines: for the same type of question, different engines returned very different lists, and no position was stable. That means the window is still open. It also means choosing the wrong partner costs more than a wasted budget — it costs you the window in which AI answers are still being formed.

Before you choose: be clear about what you are buying

"AI consulting" and "GEO" are often sold as one package. They are three different things:

  • Enterprise AI consulting solves internal problems — which processes to put on AI first, how to govern data and permissions, how to pilot and verify.
  • GEO (generative engine optimization) solves an external problem — when customers ask AI about your category, are you named accurately, cited, and recommended?
  • Traditional SEO and web development is the foundation for both — search visibility and a sound website — but it is not a substitute for either.

Decide which of the three you are missing before you shop, and make any vendor who bundles them quote each part separately.

Seven criteria that actually discriminate

1. Fluency on both sides — China and the target market

Going global is never a one-sided problem: your product narrative, Chinese-language corpus and home channels sit on one side; the target market's language, media, directories and the way customers phrase questions sit on the other. A team that only understands one side will produce work that reads wrong on the other. The test is simple: ask them which kinds of sources AI engines most often cite for questions in your category in your target market. Specifics count; generalities do not.

2. A transparent methodology: how mentions and citations are measured

GEO results must be measurable. A competent provider can state plainly: which fixed question set they sample, on which engines, how often they re-test — and they report "mentioned", "cited" and "referred visits" as separate numbers. A provider with no baseline measurement method offers results that can never be audited: without a baseline, every claimed "improvement" is unfalsifiable.

3. Technical depth they can do themselves

Being mentioned by AI starts with being crawled, indexed and correctly understood — three layers of technical work: the crawl layer (robots access, complete sitemap, clean status codes), the structure layer (page semantics and question-shaped content), and the entity layer (JSON-LD structured data, consistent facts across the web; machine-readable files such as llms.txt are optional aids). All three must be verifiable in your own server logs and audit reports. A team that only talks about content and buzz, and never touches the technical surface, cannot fix what actually blocks most companies.

4. Real compliance competence

Entering Europe, your website and AI applications run straight into GDPR, the EU AI Act and the Data Act: lawful data collection, risk classification and disclosure duties for AI features, and cross-border data arrangements all have to be designed in from the start — retrofitting them is the most expensive route. Ask directly: for our industry and product shape, what are the first gates we must pass in Europe? Judge whether you get a checklist or a slogan.

5. Evidence you can verify

Ask for data, not adjectives. Verifiable evidence includes published data reports, raw sampling records, and case work with named sources. One reference point: in our recently published AI Search Visibility of Chinese Companies Going Global (2026, First Edition), which scored the websites of 30 outbound Chinese companies, the median score was 80, and the weakest of five categories was the machine-readable surface (69.2) — structured data, sitemap and similar fundamentals. Most companies, in other words, are stuck at the basics. A provider who opens with brand buzz and skips the foundations most likely has the sequence backwards.

6. Itemised pricing (see the pricing section below)

A single lump-sum quote with no itemised deliverables cannot be compared, and cannot be accepted against. Read the three pricing layers separately, as set out below.

7. The ability to stop at each step

A sound engagement structure gives every step its own deliverable and lets you stop there: diagnose first, then decide on execution; prove one market, then decide on expansion. If diagnosis, execution and an annual retainer are welded into one contract, ask why.

Four kinds of providers, compared

DimensionLocal overseas agencyLarge consultancyChina-based teamTools only (DIY)
Understanding of the Chinese businessWeak; the client ends up educating the agencyStronger where a China practice existsStrongest — context and decision chains are nativeNot applicable; tools do not understand
Target-market corpus and channelsStrong; local media and directory access in handStrong on brand and relationships, weak on day-to-day corpus workVaries widely by firm; verify specificallyMonitoring data only; produces no mentions
European compliance (GDPR / EU AI Act)European agencies strong; others averageComplete frameworks, usually billed as separate itemsA common weak spot; examine closelyNot applicable
Pricing and contractsHigh; mostly long retainersHighest; billed by day rate and seniorityFlexible; step-by-step contracts possibleLowest subscription cost, but you staff the operations yourself
Responsiveness and communicationTime-zone and language costs are realMany layers, slow responsesSame language, no time differenceEntirely on you
Best fitCompanies with a local team that only need local executionGroup-level strategic programmes at large enterprisesFirst-time outbound companies that must spend on what mattersCompanies with an in-house growth team that only lack the dashboard

Pricing: what each layer actually buys

Quotes in this field break into three layers. When they arrive as one lump sum, ask one more question:

  • Tools layer: subscription pricing buys a monitoring dashboard — mention rates, citation sources, competitor comparisons. It answers "are we mentioned?" It does not get you mentioned.
  • Diagnostic layer: this should be a fixed price with fixed deliverables. Negency's diagnostic package, as an example: ¥19,800 paid once, delivering a human review of your audit data, a three-competitor comparison, a 90-day roadmap and a walkthrough session — with no obligation to continue. Treat this layer as a litmus test: a provider who will not fix a price for diagnosis will not price execution transparently either.
  • Execution layer: quoted by scope and by number of target markets. Every company starts from a different position, so this layer cannot be standardised; what you can compare is not the total, but the itemised deliverables and how each is verified.

The rational starting point is a free audit (machine-generated score): get your own baseline score and itemised issues before discussing any price.

Red flags: any one of these should end the conversation

  • "Guaranteed #1 in ChatGPT" or "guaranteed recommendations" in a named engine — engines do not sell that slot, and nobody can guarantee it.
  • Paid mentions, fabricated citations, bulk low-quality backlinks — these plant a mine under your brand; when engines identify them, the long-term damage is yours.
  • A total price with no itemised deliverables and no acceptance criteria.
  • One client case presented as a general effect, with a refusal to show verifiable data sources.
  • No baseline measurement — or a refusal to disclose the question set used for measurement.

Where Negency sits in this table

Stated plainly: Negency is a Sweden-based AI consulting and GEO company whose team works long-term on both the China and Europe sides; its home ground is precisely the China-to-Europe bridge. We do two things: GEO growth — getting companies mentioned accurately and cited in AI answers — and enterprise AI consulting and agent deployment, including compliance design for GDPR and the EU AI Act. The verifiable evidence is our published AI Search Visibility of Chinese Companies Going Global (2026, First Edition), with itemised scores for all 30 companies in the open. Every criterion in this guide applies to us as much as to anyone else.

Next step

Spend fifteen minutes on a free AI visibility audit and get your company's baseline score; then take this checklist to any provider you talk to. For the peer benchmark, the full report is here; to have your issues broken down item by item, the fixed-price ¥19,800 diagnostic is the low-risk first step.