Coaley Peak GEO Research Centre Multilingual · Multi-country
Be the answer. International GEO, AEO and AI SEO for results in foreign markets.
We are a multilingual, multi-country research centre for generative engine optimisation, answer engine optimisation, AI optimisation and AI SEO, and our speciality is foreign markets. We study how AI answers are composed in each country and language, implement what gets cited there, and train the SEO teams who carry it forward.
Three complimentary hours for selected teams, at our discretion.
15+
years of SEO experience leading the research centre
28
engines and assistants we target
16
of them built outside the UK and US
3
complimentary hours for selected teams
Why this research centre exists
Search used to return ten blue links. Now it returns one answer, composed by a model from what it retrieved a second ago and what it learned months before. In Shanghai that answer often comes from Baidu or Doubao, in Seoul from Naver, in Tokyo from Yahoo! JAPAN, each in its own language. Most GEO still targets one engine in one language. We built a research centre for every other market.
International
Results in foreign markets are our core speciality
Most GEO advice assumes Google and ChatGPT. In much of the world the answer comes from somewhere else, each engine trusts different sources, and an English strategy translated word for word rarely gets cited. Getting results in foreign markets is the core of our research.
China
Baidu
Rewards Simplified Chinese content, mainland hosting with an ICP licence, and its own properties, Baidu Baike and Baijiahao, as entity and citation sources.
South Korea
Naver
Leans on its own ecosystem, Naver Blog, Cafe and Knowledge iN, so visibility is built inside Naver as much as on your website.
Japan
Yahoo! JAPAN
Runs on Google's index, but Japanese searchers phrase intent in ways a translated English keyword set rarely matches, and its own Q&A and news properties surround the results.
Czech Republic
Seznam
Crawls with its own bot, SeznamBot, and draws local business data from Firmy.cz, in a market where Google is not the only gatekeeper.
How we get results abroad
- Native-language prompt research: what people in each market ask, in their own words, not a translated keyword list.
- Local entity sources: Baidu Baike, Naver's encyclopedia, local Wikipedias and Wikidata labels, all describing you the same way.
- Technical access for each market: crawler permissions, hosting, rendering and latency, including licensing where it applies.
- Transcreation over translation, so the passage an engine lifts reads as if it were written there.
- Measurement country by country: prompt panels run in each language and location, so results are reported market by market.
Foreign LLMs
Foreign large language models
Which assistant your customers use depends on where they live. ChatGPT is not officially available in mainland China, where people ask Doubao, DeepSeek, Qwen, Kimi, Yuanbao or Baidu's ERNIE instead. Korea has Naver's HyperCLOVA X, France has Mistral, and the Gulf has Arabic-first models such as Jais and ALLaM.
Foreign models matter at home too. Open-weight models from China and Europe, Qwen, DeepSeek and Mistral among them, sit underneath thousands of apps and agents worldwide. When those apps answer without searching, what the underlying model learned is all that counts.
We test each model in its own language, map the sources its answers draw on, and track how each new release changes what it says about you.
- Doubao
- DeepSeek
- Qwen
- Kimi
- Tencent Yuanbao
- ERNIE
- HyperCLOVA X
- Mistral
- Jais
- ALLaM
Engines
Every engine that composes an answer, in every market
If it composes an answer, we can optimise for it: the engines and assistants that lead foreign markets, global AI search, and the open-weight models running inside thousands of other products.
China
- Baidu
- Doubao
- DeepSeek
- Qwen
- Kimi
- Tencent Yuanbao
South Korea
- Naver
- Daum
Japan
- Yahoo! JAPAN
Europe
- Seznam
- Mistral Le Chat
- Qwant
- Ecosia
Middle East
- Jais
- ALLaM
South-east Asia
- Cốc Cốc
Global AI search and assistants
- Google AI Overviews
- Google AI Mode
- Gemini
- ChatGPT
- Microsoft Copilot
- Bing
- Perplexity
- Claude
- Meta AI
- Grok
- DuckDuckGo
- Brave Search
Engine and model names are trademarks of their respective owners. Coaley Peak is independent and is not affiliated with, or endorsed by, any provider named on this page.
Definitions
GEO, AEO, AIO and AI SEO, defined
The labels overlap, and the industry uses them loosely. This is how we use them.
- GEOGenerative engine optimisation
- Making a brand the source that generative engines retrieve, trust and cite when they compose an answer, from ChatGPT and Perplexity to Google's AI Overviews and their equivalents abroad.
- AEOAnswer engine optimisation
- Structuring content so that any system returning a single answer, from a featured snippet to a voice assistant, can lift it accurately and attribute it to you.
- AIOAI optimisation
- Checking what AI systems actually say about a brand, then fixing what is wrong at the source, wherever that source lives.
- AI SEOSearch engine optimisation for AI
- Technical SEO for a web read by machines: crawl access, rendering, structured data and entity work that serve search crawlers and AI agents alike.
- Base model targetingOur specialist service
- Influencing what a language model knows about a brand before it searches anything, by improving the data that pre-training draws on and measuring the result with retrieval switched off.
Research centre
A research centre, not a reseller of opinions
Most GEO advice is anecdote. Ours is measured: every recommendation traces back to a test you can inspect, with the prompts, markets, languages, models and raw transcripts attached.
Prompt panels
Fixed sets of real prompts, run on a schedule across engines, languages and locations, so share of answer is a number you can track.
Citation forensics
Which sources each engine cites, how often and in what position, so we know which third-party pages actually move an answer.
Controlled experiments
Structure, schema, phrasing and freshness tested against control pages before anything reaches a client's site.
AI crawler analytics
Server logs read for the bots that gather retrieval and training data: GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Baiduspider, Yeti, SeznamBot, Bytespider and more.
Base model probing
Base and open-weight models queried with retrieval switched off, to measure what they already believe about a brand and its competitors.
Release tracking
New models re-tested against our panels on release, so a shift in a client's visibility reaches you from us, not from their customers.
Specialist service
Base model targeting: GEO below the retrieval layer
Base model targeting influences what a language model knows about a brand before it searches anything. Most GEO works only on retrieval, getting cited when an assistant looks something up. We also work on the model's memory: the knowledge fixed in pre-training, which shapes every answer given without a search and frames the ones given with one.
That memory is uneven across languages: a brand a model knows well in English can be missing, or wrong, in Korean or Mandarin. Few agencies know this layer exists, and fewer can measure it. It takes data science rather than content alone: running open-weight models, probing them at scale and reading the statistics honestly.
Layer 1 · Pre-training
The base model
Trillions of tokens of web pages, books, code and reference data, compressed into what the model knows by default and fixed at a knowledge cut-off.
Our work: We probe what base and open-weight models already associate with a brand, in each market's language, and trace the training sources behind it.
Layer 2 · Post-training
Tuning
Instruction and preference tuning decide how the model speaks, what it hedges on and which kinds of source it trusts.
Our work: We map those preferences model by model, so content is shaped for what each one rewards.
Layer 3 · Retrieval
Live search
Searching the web at the moment of the question: the layer most GEO works on, and the only one most agencies measure.
Our work: Classic GEO and AEO, across every engine that retrieves, in every market we cover.
Layer 4 · The answer
What people read
Memory and retrieval blended into one response, with or without a citation.
Our work: We measure the blend, so you know which layer moved when the answer changed.
How base model targeting works
Probe
Hundreds of phrasings in each target language, retrieval switched off, across base and open-weight models. On open-weight models we read the token probabilities directly.
Trace
Where pre-training data comes from: Common Crawl, Wikipedia and Wikidata, reputable publications, documentation and public datasets. We find where a brand is missing, inconsistent or out of date.
Shape
Accurate, consistent and corroborated information placed where it legitimately belongs, with training crawlers able to reach it, if that is your policy.
Track
Each new model generation re-probed, to show whether a brand's representation improved, held or slipped.
No data poisoning, no synthetic spam, no hidden prompts. If it would embarrass you in a model's training data, we will not put it there.
Services
What we offer SEO agencies and in-house teams
Built for SEO agencies and in-house teams taking brands into foreign markets, or growing them there. Use us for the research, the implementation, the training, or all three.
- 01
International visibility audits
AI visibility audits by country, language and engine, share-of-answer benchmarks against local competitors, and base model probes in each language. Transcripts always included.
- 02
Market entry and expansion
GEO for a market you are entering or growing in: native-language prompt research, local entity sources, transcreated content and technical access, country by country.
- 03
Implementation at scale
Multi-market, multi-language rollouts: programmatic entity and structured data work across thousands of URLs, AI crawler log pipelines and measurement per market.
- 04
Implementation for single sites
One site or brand, done properly in every market it serves: entity groundwork, answer-first content, technical access and a measured baseline.
- 05
Tactics and playbooks
For teams who would rather do it themselves: our market playbooks for engines such as Baidu, Naver, Yahoo! JAPAN and Seznam, with templates, QA checklists and prompt panels.
- 06
Training
GEO, AEO and AI SEO training for SEO teams, from fundamentals to international GEO and base model targeting. From £450 + VAT for a half day, in person, with our satisfaction guarantee. Training details
Small scale or large scale
The principles are the same at any size. What changes is how the work is built and measured.
| Aspect | One site | Large estates |
|---|---|---|
| Scope | One brand, one or two markets | Many sites, brands, markets and languages |
| Entity work | Hand-built entity graph and structured data | Entity data and schema generated from your catalogue or CMS |
| Content | Priority pages rebuilt answer-first | Templates and pipelines that restructure content at scale, with human QA |
| Measurement | Prompt panel baseline and scheduled re-runs | Logs, prompt panels and analytics in one data pipeline |
| Foreign markets | Your priority market first | Market by market, with model release tracking |
Running GEO for a business rather than an agency? See our GEO programme for dental practices.
Research centre lead
Led by Stephen Grindley
Stephen Grindley leads the research centre. He has more than 15 years in SEO, alongside a background in data science and data engineering, and spent over a decade building data systems for businesses operating across JAPAC, EMEA, LATAM and NAMER. That is why our GEO work runs like a research programme rather than a content calendar. He founded Coaley Peak in 2019.
Ethan Clarke-WoosterBusiness Development Executive
Ethan is your first point of contact, and you will find both of them on the live chat on this page.
3 hours GEO consultancy for free
For selected teams
Test our team: three complimentary hours
For selected teams, we are offering three hours of our time at no charge, so you can test our team and capabilities on a real problem. Use them for a visibility check in a foreign market you are entering, a base model probe on a brand in another language, an audit across engines, or a working session with your SEO team.
To apply, tell us who you are, what you want to test, and which engines or markets matter to you.
Please note: Complimentary hours are offered to selected organisations only. Coaley Peak chooses who receives them at its sole discretion, and applying does not guarantee a place. There is no obligation to buy anything afterwards, and we may change or withdraw this offer at any time.
FAQ
Questions SEO teams ask us
Anything else? Ask Stephen and Ethan on the live chat.
What is the difference between GEO, AEO and AI SEO?
They overlap. AI SEO keeps a site crawlable, understood and trusted by search engines and AI agents. AEO shapes content so any single-answer system can lift and attribute it. GEO goes further, making a brand the source that generative engines retrieve and cite. We work on all three, and on base model targeting beneath them.
Which countries and languages can you target?
Any market where an engine or assistant composes answers. We specialise in markets where Google and ChatGPT are not the whole story: China, South Korea, Japan, the Czech Republic and the rest of Europe, the Gulf and South-east Asia. Every market is researched in its own language, never through a translated English strategy.
What is base model targeting?
Influencing what a language model knows about a brand from pre-training, rather than only what it finds when it searches. We measure it by probing base and open-weight models with retrieval switched off, then improve the sources that training data is drawn from.
Which search engines and AI assistants do you cover?
Baidu, Doubao, DeepSeek, Qwen, Kimi and Yuanbao in China, Naver and Daum in South Korea, Yahoo! JAPAN, Seznam, Mistral, Jais, ALLaM and more, plus Google's AI Overviews and AI Mode, Gemini, ChatGPT, Copilot, Bing, Perplexity, Claude, Meta AI and Grok: 28 in all, listed on this page.
Can you guarantee an AI assistant will recommend a brand?
No, and nobody honestly can: no one controls a model's output. We control whether a brand is findable, resolvable and quotable, in retrieval and in training data, and we measure the effect against a baseline taken before work starts.
How do you measure GEO results?
Fixed prompt panels, re-run on a schedule across engines, languages and locations, record whether a brand is named, cited and described accurately. Server logs show AI crawler activity, and analytics show AI-referred traffic.
How much does GEO training cost?
Training runs at our standard rates, in person, for the whole group: £450 + VAT for a half day, or £787.50 + VAT per day. If you are not entirely satisfied, tell us within 14 days of the session and we refund the training fee. Implementation is scoped and quoted per engagement.
Who can apply for the three complimentary hours?
Any SEO agency, in-house SEO team or consultant can apply. We choose who receives them at our discretion, based on fit and capacity, and applying does not guarantee a place.
Want to work with us or for us?
Work with us
Become a client
- Stephen GrindleyResearch centre lead
Ethan Clarke-WoosterBusiness Development Executive
Tell us which markets you want to win. We will show you where you stand in each, in its own language, before we propose anything. Stephen and Ethan answer the live chat on this page, or call us free on 0800 494 7725.
Work for us
Join the research centre
We like to hear from multilingual SEO specialists, native-language researchers and data scientists who want to get results in foreign markets.
How to apply
- See our open roles and apply online, or email your CV to careers@coaleypeak.co.uk if none fits yet.
- Tell us, in your own words, which markets and languages you know.
- We acknowledge every application within five working days, and our process includes a short exercise we pay you for.
Reviewed by Stephen Grindley, research centre lead. Last updated .