AI SEO Agency: What They Actually Do When Search Runs on AI Answers
An AI answer in Google, ChatGPT or Perplexity typically cites 3 to 8 sources. Being one of them is now a separate discipline from ranking at position 3, because the answer is assembled above the results and many users never scroll past it. An AI SEO agency exists to make your pages the ones that get quoted, and to keep the classic rankings that feed those citations.
This is not a replacement for SEO. Every study of AI citations so far points the same way: the pages that get cited are overwhelmingly pages that already rank on page one. AI SEO is SEO plus a specific layer of work on top.
What actually changed
Three things, in order of business impact:
- The answer moved above the link. Informational queries increasingly end in the answer box. Click-through on informational keywords is down; click-through on commercial and transactional keywords is far more resilient, because people buying still want to see the vendor.
- Citation is the new impression. Being named in an AI answer, even without a click, puts you on the shortlist. Users then search your brand name directly, which is why brand-search volume is the cleanest early signal that AI SEO is working.
- Question phrasing replaced keyword phrasing. People type full questions into an AI, not two-word keywords. Content structured as a clear question with a direct answer underneath gets extracted; content that buries the answer in paragraph six does not.
The seven workstreams of a real AI SEO engagement
1. Extractable content structure. Every important page gets question-shaped headings with the answer in the first 40-60 words underneath, before the context. This single change is the highest-leverage AI SEO work there is, because extraction models read the first sentences under a heading.
2. Concrete, checkable facts. AI systems preferentially cite pages containing specific numbers, prices, timeframes and named tools, because those are the parts that make a useful answer. "Affordable and fast" is uncitable. "EUR 2,000 to EUR 6,000, live in 2-4 weeks" gets quoted.
3. Structured data. FAQPage, Article, Service, Organization and Breadcrumb schema, correctly implemented, so both crawlers and AI systems can parse what a page is and who published it. FAQPage schema on your buyer questions is the fastest win.
4. Entity and brand consistency. Same company name, description, founding data and service list across your site, Google Business Profile, LinkedIn, directories and Wikipedia-class sources. AI models build an entity picture from consensus across sources; contradictions weaken it.
5. Crawlability for AI crawlers. GPTBot, ClaudeBot, PerplexityBot and Google-Extended each have their own access rules. Many sites accidentally block the crawlers whose citations they want. This includes an llms.txt file and a robots.txt policy you have actually decided, not inherited.
6. Buyer-question coverage. Content built directly against the questions buyers ask an AI: what it costs, how to choose, X versus Y, what to ask before signing. These pages convert far better than top-of-funnel volume plays, because the person asking them is mid-decision.
7. Classic technical and authority SEO. Core Web Vitals, internal linking, and links from sources AI systems trust. Skipping this and doing only "GEO" is the most common way agencies waste a budget: without page-one rankings there is nothing for the answer engine to cite.
How to measure it
Ranking reports do not capture AI visibility. Track these five instead:
Citations in AI answers. Source: a tracked prompt set of 20-40 buyer questions, checked monthly. Good looks like: a rising share of prompts where you appear.
Referral traffic from AI tools. Source: analytics, referrers chatgpt.com, perplexity.ai, gemini. Good looks like: growing, and converting better than your organic average.
Brand search volume. Source: Search Console, queries containing your name. Good looks like: steady growth. This is the clearest proof of AI exposure.
Impressions on question queries. Source: Search Console, queries starting who, what, how or why. Good looks like: rising alongside citations.
Leads who mention AI research. Source: a field on your own intake form. Good looks like: a real number, not an anecdote.
Set up the prompt set in month one, or you will have no baseline and no way to prove anything in month six.
What it costs
- One-off AI-readiness audit and fixes: EUR 1,200 - 3,500. Structure, schema, crawler access, entity consistency across your main pages.
- Ongoing programme: EUR 800 - 3,500 a month. Content against buyer questions, technical SEO, authority building, and monthly citation reporting. The range is driven by content volume and how competitive your market is.
- Content only: EUR 250 - 600 per substantial article built to be citable.
Typical time to first measurable movement: 6-10 weeks for citations on low-competition buyer questions, 4-6 months for competitive commercial terms.
Five warning signs when hiring
- They promise a ranking or a citation. Nobody controls an answer engine's output. Anyone guaranteeing placement is selling certainty they do not have.
- They sell GEO with no technical SEO. Citations come from pages that rank. There is no shortcut around that.
- No prompt-set baseline. Without a tracked set of questions measured from month one, "AI visibility improved" is unfalsifiable.
- Volume-first content plans. 40 thin articles a month is exactly the pattern search engines now penalise as scaled content abuse.
- No access to your own Search Console data in the plan. Any serious plan starts from your real queries and impressions, not from a generic keyword tool export.
Where to start if you do it yourself
This week: pick your ten highest-intent buyer questions, write a direct 40-word answer under a question-shaped heading for each on the relevant service page, add FAQPage schema, and check your robots.txt does not block the AI crawlers. That is a day of work and it is the majority of the available upside.