Understanding and using AI for an effective international SEO strategy - A talk at the SEO Camp Day Strasbourg 2023

→ Link to the slides in French 🇫🇷

In February 2023, I stood in front of a room of French SEO practitioners at the first SEO Camp Day Strasbourg and asked a question the industry was dancing around: what can AI tools actually do in international SEO, and where are they performing confidence without competence?

My audience were practitioners: curious, slightly cautious, and smart enough to spot hype. Rather than adding to the noise around ChatGPT, I tested AI against the specific demands of multi-market SEO work: tasks where language, cultural nuance, and structural precision all matter at once. What follows are the key takeaways from the session, the resources I cited with annotations, and an honest update from 2026.

What I hope people took away from the talk

AI processes patterns, not meaning, and that gap matters in international SEO. The Chinese Room thought experiment is the right frame: a person inside a room follows rules to process Chinese symbols without understanding the language. AI does the same. For international SEO, this matters because cultural nuance is meaning. Any task that requires understanding why a German-speaking audience searches differently from a French-speaking one sits outside what current AI handles reliably.

Hreflang generation is a legitimate use case. Generating hreflang tag sets from a structured input (list of URLs, language/country combinations) is exactly the kind of mechanical, pattern-based task where AI performs well. The output format is predictable, the validation rules are known, and errors are easy to catch. Use it to rough out the implementation, then verify against Google's specification before deploying.

Schema markup generation works for the same reason. FAQ schema, Product schema, LocalBusiness — these follow strict structures with defined properties. Feeding AI your question-and-answer pairs and asking for JSON-LD output cuts most of the manual work on a repetitive task. The result needs review, but the time saving is real. Always validate through Google's Rich Results Test before implementing.

Translation is unreliable, especially for anything with texture. I tested DeepL, Google Translate, and ChatGPT on the same passage of nuanced English prose. All three lost something — particularly figurative speech and culturally specific connotations. For international SEO, where keyword meaning differs across markets (French Canadian gommevs. French chewing-gum, for instance), machine translation gives linguistic proximity, not market relevance. Use it as a starting point, then have a native speaker review.

Cultural content adaptation works across large differences, not subtle ones. Asking AI to generate content ideas for a gardening blog in Finland versus Morocco produces meaningfully different outputs — climate, local customs, regional specifics come through. Asking it to adapt content between France and Belgium, or between UK and Australian English, produces near-identical results. The smaller the cultural gap, the less AI adds over just writing for the market directly.

Alt tag generation at scale is a strong use case. Danny Richman's GPT-3 alt tag generator combined the OpenAI API, EveryPixel for image analysis, and RapidAPI for orchestration. For large international sites with thousands of product images, this kind of tooling saves real time at acceptable quality. The "under the hood" slide from the talk showed the actual prompt structure — it specifies output length, excludes decorative language, and includes keyword context.

Building lightweight tools is within reach. The most practical part of the talk was showing Python and Google Apps Script in action for repetitive SEO tasks: extracting images without alt text from a Screaming Frog export, automating monthly report delivery to clients. These aren't complex — but they save time every single month and remove recurring tasks from the mental load entirely.

Resources from the talk

What is generative AI? — McKinsey This is where the working definition in the talk came from. It covers the distinction between generative models (text, image, audio, code) and more traditional AI approaches without requiring a technical background. Worth sharing with a client or marketing team trying to understand what these tools actually are before deciding whether to use them.

GPT-3 explained — Science Focus GPT-3's 175 billion parameters were the clearest illustration at the time of the scale difference between large language models and anything that came before. This article explains the architecture accessibly. The parameter count comparison in the talk (GPT-3 versus the human brain's estimated synapses) made the scope of the technology concrete in a way that pure definitions don't.

Chinese room thought experiment — Wikipedia (FR) Searle's thought experiment is the most useful conceptual frame for understanding what AI is not doing. The room processes symbols by rule without understanding their meaning — which is precisely the distinction that matters when cultural nuance is part of the work. If you want the English version, search "Chinese room Searle" — the concept is the same in any language, which is appropriate given the context.

ChatGPT for SEO — Aleyda Solis Aleyda Solis was among the first practitioners to document specific prompt patterns for SEO tasks in a structured, replicable way: hreflang generation, schema markup, meta descriptions. The examples in the hreflang and FAQ schema sections of the talk drew directly from this resource. It remains a useful starting library for anyone building their own prompt collection for technical international SEO work.

Generate alt tags with ChatGPT — Danny Richman Danny Richman built a working alt tag generator by combining the OpenAI API, EveryPixel for image recognition, and RapidAPI. The talk showed the actual prompt structure: generate a descriptive alt tag under 16 words, from image URL plus keyword context, excluding decorative language like "illustration" or "wallpaper." Worth reading if you want to understand how to replicate something similar for a client site with a large image inventory.

ChatGPT: Friend or Foe? — The Recipe for SEO Success (Kate Toon) Kate Toon's podcast episode covers the practical and strategic questions around AI in SEO from the perspective of a grounded generalist practitioner — a useful counterweight to the breathless coverage that was everywhere in early 2023. Kate later invited me onto the show to discuss SEO longevity, which you can listen to here.

My take in 2026

The core argument holds: AI handles structured, rule-based international SEO tasks well and breaks down where cultural understanding is required. What has shifted is the baseline. Models are far stronger now, real-time web access has resolved the "no recent data" limitation for most tasks, and the barriers to building lightweight tools have dropped considerably — what required a Python script and an API key in 2023 takes ten minutes in a tool like Claude Artifacts. The toolbox has expanded significantly. What hasn't changed: AI still doesn't understand why German users search differently from French ones, or why the same product concept needs completely different framing in Quebec versus Paris. That judgment still sits with the practitioner.