AI SEO agency: what it should actually deliver

Why every agency became an “AI agency” this year

Google search itself has shifted: generated answers (AI Overviews and AI Mode, rolled out in France on 22 July 2026, after most other markets) capture part of the clicks that used to reach websites. Businesses see it in their analytics, look for an answer, and agencies adjusted their storefront: the word “AI” appeared on homepages much faster than in working methods.

Telling the two apart takes fifteen minutes of conversation. An agency that has genuinely industrialised its SEO with AI can show you three things: its engine running, its review pipeline, and its costs. An agency that “added AI” will show you slides.

What an AI SEO engine concretely does

Classic agency SEO moves in batches: an audit, a monthly report, a call, recommendations, then next month it starts again. An AI-assisted SEO engine moves continuously, across four functions:

• Opportunity detection: a regular Search Console scan crosses your actual positions with a keyword objectives registry. Queries within striking distance, pages slipping, titles seen but never clicked: the week’s work comes out of data, not intuition.
• Production: models draft content structured to be cited (a direct answer up top, sourced facts, planned internal linking), in several languages if your market requires it.
• Technical audit: the robots.txt actually served, the sitemap, hreflang, access for AI crawlers (GPTBot, ClaudeBot), and titles rewritten when impressions are there but clicks are not.
• Refreshing: content losing positions gets diagnosed and brought back up to date, facts re-checked and re-sourced, rather than left to slide.

We can describe this mechanism precisely because it is our own engine’s: it has scanned our Search Console every evening since 25 August 2026, and every draft it produces goes through human review before publication. This article is one of its products.

Human review is not optional

The one point never to concede: no generated content should ever be published without human review. We apply this rule because we learned it the expensive way: on one of our sites, a model invented three real-estate projects inside an article, which went live and had to be urgently unpublished. False content costs more than ten unpublished drafts, and that is truer than ever now that generative engines repeat what they read.

A serious AI SEO agency tells you who reviews, at which step, and what blocks a publication. If the answer is “AI is very reliable now”, the conversation can end there.

The 5 questions that expose a fake AI SEO agency

1. “Show me your engine on your own site.” An agency selling SEO industrialisation must apply it to itself, and be able to show dates, reports, and a history.
2. “Who reviews, and what blocks a publication?” The right answer describes a pipeline, with a human in it and written rules.
3. “What happens to my site if we stop?” Content, data and access must remain yours. Twelve-month lock-ins with vague reversibility are a signal, not a norm.
4. “What does it cost to run, and how do I see it?” AI has crushed production costs: a retainer whose price structure has not moved should explain where the value goes (strategy, review, technical work); we break down the market’s real ranges in our article on SEO agency pricing in 2026.
5. “Where does my data go?” Which models, which providers, which policy for sensitive data. An agency that masters its craft has a written answer, per type of data.

What it changes for a small business

For a 5 to 50 people company, the engine mostly changes the rhythm and where the budget goes. The rhythm: small passes every week instead of one big quarterly project, so content follows your customers’ real questions rather than the agency’s calendar. The budget: it no longer goes mostly into production by the kilo, it goes into what makes the difference, keyword strategy, expert review, technical work, and proof measured in your own Search Console.

Visibility in generative engines (being cited by ChatGPT or Perplexity, known as generative engine optimization) runs on the same engine: content structured to be cited serves both. That is the subject of this category’s pillar page, SEO and AI visibility.

Frequently asked questions

Is an AI SEO agency cheaper than a classic one? Production costs less, yes. But the price of a retainer depends mostly on scope and review standards. Beware of a very low quote with no review pipeline: you are paying little for content that Google and generative engines are learning to ignore.

Can AI run SEO on its own? No. It detects, produces and monitors very well. It can neither guarantee a fact nor decide what your business should say. Those last two are precisely what builds long-term visibility.

Does SEO still matter with AI-generated answers? Yes, twice over: generative engines cite the sources they consider reliable, and those sources are overwhelmingly the ones doing good SEO. The work changes shape, not usefulness.

How do I check an agency does what it claims? Ask to see its engine applied to itself, its dated reports, and one example of a draft rejected in review. The rejection is the best proof: a pipeline that never rejects anything does not review.