AI visibility is a new layer on top of SEO. It doesn't replace it — it complements it. Sites that understand this first and invest resources in GEO will, in 12 months, be the defining names of their category inside AI answers. Sites that „wait and see“ will pay several times over for the same position in 2027.
Why we're writing about this now?
In January 2026 we redesigned the site for a Croatian FinTech. We tracked the usual: organic traffic, conversions, CAC. All growing. Then their COO came in with a number that stopped us cold: 31% of new customers from the prior three months, when asked how they heard about us, said they'd seen us in a ChatGPT answer.
We weren't tracking it. We didn't know how. We were excellently SEO-optimized and simultaneously invisible to half of our customers, who were deciding in a place we weren't looking. That's the moment we stopped calling this „SEO“ and started talking about AI visibility as a separate discipline.
The definition
AI visibility (also called GEO - Generative Engine Optimization) is a measure of how often your brand, products, or content are mentioned, cited, or recommended by AI systems (ChatGPT, Claude, Perplexity, Gemini, Google AI Overviews) when users ask questions relevant to your business.
The three key words in that definition are mentioned, cited, and recommended. Those are three different things and all three count. A mention means the AI knows you exist. A citation means it uses you as a source with a link. A recommendation means it actively suggests you as the best option ("if you're looking for X, consider Y and Z, but the strongest is W").
| Dimension | SEO (classic) | AI visibility |
|---|---|---|
| What it measures | Rank in Google results (top 10 for a query). | Citation frequency inside AI answers - with or without a link. |
| User action | Clicks a link → lands on your site. | Reads the answer → often doesn't click anything. |
| What the algorithm prefers | Backlinks, on-page optimization, keywords, speed. | Structured content, clear definitions, brand mentions in quality sources, original data. |
| Update frequency | Google indexes continuously. | LLMs train in waves (RAG vs training); citations change within days. |
| How it's measured | Google Search Console, SERP trackers. | Manual querying + logging, or tooling (Profound, Athena, custom scripts). |
| Winning looks like | Top 3 organic rank. | Your brand mentioned in the first answer for your top 10 questions. |
The most important row in the table is the second one: when the user no longer clicks, classic SEO is only watching part of the game. If ChatGPT recommends a competitor, the buyer goes directly there. Your Google rank, however good, played no part in that decision.
Why now? What changed in 18 months?
It wasn't gradual. Three things converged.
1. ChatGPT got web browsing and source citations
Up to mid-2024, ChatGPT „knew“ things only from training. Now it browses live, cites sources, and is the primary answer for millions of people. Same for Claude (via Claude Search), Perplexity (from day one) and Gemini.
2. Google AI Overviews took over the top of the SERP
AI Overviews (formerly „Search Generative Experience“) now appears on roughly 47% of commercial queries. Meaning, even when you rank in the top 3 of the classic SERP, if you aren't cited in the AI Overview at the top, half the clicks are eaten by it.
3. Buyer behaviour changed
Our internal survey of 412 respondents in Q1 2026: 38% of B2B buyers and 31% of e-commerce buyers in Croatia and the region „check with AI before clicking any link“. That's 6× more than Q1 2024.
Your site is no longer the first touch with the prospect. The AI answer is. If AI doesn't know you, or doesn't consider you relevant — you lose the prospect before they ever see your homepage.
How AI systems decide whom to cite
Different LLMs have different mechanisms - RAG (retrieval-augmented generation), grounded responses through Bing/Google indexes, training on the public web, but five factors recur.
Factor 1: Structured content
LLMs understand structure. A page with a clearly tagged FAQ section (FAQPage schema), a defining first paragraph and a logical H2/H3 hierarchy is „easier to lift“ than a wall of text. Schema.org markup is to AI what backlinks were to early Google.
Factor 2: Clear, quotable definitions
When the user asks „what is X“, AI looks for a page that gives a direct answer within the first 100 words — a one-sentence definition. Sites that „first talk about themselves, then talk about the problem“ lose. Sites that answer the question in the first paragraph get cited.
Factor 3: Mentions on other quality sources
LLMs „know“ about you from the internet — not just from your site. A mention in an industry publication, podcast, interview, guest post, GitHub repo, Wikipedia entry, Reddit thread (yes, seriously) — all of it builds your brand's „weight“ inside AI systems.
Factor 4: Original data and research
LLMs strongly prefer citations to sources that are the origin of data, not compilers. If you publish original research („State of Croatian Shopify stores 2026“), internal benchmarking, or concrete numbers from your projects — you'll be cited as a primary source, not „also mentioned“.
Factor 5: Topical authority
You don't get cited from a page about an unrelated topic. If you write about everything-everywhere, AI won't see you as a source for anything. If you write about 12–20 connected topics within one cluster, you become „that site about X“ — and AI returns to you every time the user asks about X.
Who's already getting cited? A real example.
In Q1 2026 we measured AI visibility for five Croatian and regional e-commerce brands in the „natural cosmetics“ category. We asked ChatGPT 30 different questions real customers ask („best cream for dry skin“, „natural alternative to...“, „what to buy for the first purchase“, etc.). We logged how often each brand was mentioned in the first five lines of the answer.
| Brand | Citations (of 30) | Why? |
|---|---|---|
| Brand A (local) | 22 | 12 deep „what is / how to choose“ articles. Wikipedia entry. 3 guest interviews in industry publications. |
| Brand B (regional) | 14 | Good SEO, but no original research. Cited only on broad topics. |
| Brand C (DTC focus) | 3 | Beautiful brand site, but zero content. AI „doesn't know“ what they do beyond the name. |
| Brand D (category leader) | 8 | Big SEO budget, mass-produced low-quality content. AI flags it as a spam source. |
| Brand E (new entrant) | 11 | 12-month-old site with 8 very deep articles. Consistent structure. Schema everywhere. |
What this tells us?
Brand A wasn't the biggest. Didn't have the biggest marketing budget. But it won AI visibility because it did three things AI rewards: deep „how to choose“ articles, authoritative external signals (Wikipedia, interviews) and consistent structure. Brand D proves that SEO mass doesn't win, AI recognises quality.
How to measure AI visibility? Three practical methods.
Method 1: Manual querying (free, 2 hours / month)
Simplest: build a list of 30–60 questions your customers ask AI. Ask each in ChatGPT, Perplexity, Claude and Google AI Overviews. Log: mentioned, cited with link, recommended, or omitted. Repeat monthly. That's your personal Citation Report.
Method 2: Tooling (Profound, Athena, Otterly)
Specialised SaaS products that automate Method 1 — ask hundreds of questions daily, track citations, give a dashboard. Roughly €200–800/month depending on scope. Worth it for brands that already prioritise AI visibility and want weekly instead of monthly data.
Method 3: Custom scripts (for serious teams)
API access to OpenAI, Anthropic, Perplexity. You write a script that asks questions, parses answers, logs mentions and writes to a database. Open, flexible, but needs dev resources. We do this for all AI visibility clients — but it isn't necessary for most.
90-day plan: what to start on Monday
Weeks 1–4: Audit
- Map 30–60 questions your customers actually ask AI. Survey 10 current customers, review your FAQ, study competitors.
- Ask each question in ChatGPT, Perplexity, Claude, and Google AI Overviews. Log: where you are, where competitors are, where no one is.
- Baseline number: what percentage of 60 questions mention you? The 90-day goal is +25 percentage points.
Weeks 5–10: Build
- Implement schema.org markup: Organization, FAQPage on all FAQ sections, Article on all blog posts, BreadcrumbList globally.
- Write 6–10 deep articles that directly answer the questions where you have 0% visibility. Format: TL;DR at the top, clear H2/H3 structure, FAQ at the bottom, 1500+ words.
- Identify 3–5 quality external sources where you can be mentioned: industry publications, podcasts, Wikipedia (if your category has an entry).
- Publish 1 original research piece with concrete numbers from your data (anonymised). AI loves primary-source data.
Weeks 11–13: Measure and optimise
- Repeat the Audit (week 1) with the same 60 questions. Compare numbers. Growth should be visible.
- Identify the top 3 questions where you're now visible, and the top 3 where you aren't — and write 3 more articles targeting the gaps.
- Set a monthly rhythm: Citation Report on the 1st of each month, summary into the CRM or Slack so everyone sees.
Five mistakes we see every week
1. Mass-generating content with AI tools
Ironically, AI systems are excellent at recognising low-value AI-generated content and treating it as a spam signal. Brands that „flood“ their blog with 50 ChatGPT-generated articles per month typically see AI visibility drop.
2. Assuming AI visibility = SEO
There's overlap, but it isn't the same. A page ranking 4th on Google may be cited in 18 of 30 AI questions. A page ranking 2nd may be cited 0 times. Speed, structure and definition clarity are often more important than raw rank.
3. Measuring only „mentions“ without context
AI can mention you as „a bad example“ or „not for you“. Don't count every mention as a win. Log the tone of the answer — positively recommended, neutral, or with a warning.
4. Ignoring Wikipedia and Wikidata
Many LLMs heavily rely on Wikipedia and Wikidata for foundational brand facts. If your category has a Wikipedia article and you aren't in it, that's one of the cheapest interventions — and most brands skip it.
5. No FAQ section with schema markup
FAQPage schema on a FAQ section dramatically increases your chance of being directly cited. Literally a technical lever most marketing teams have on a „someday“ list — and which takes 30 minutes of dev time to activate.
What now?
If your team hasn't started measuring AI visibility yet — start this week. An hour of manual testing through ChatGPT and Perplexity on 10 key questions will give you an honest baseline. If you're in the top 3 of the answer for most — great, build on what you have. If you aren't — you have space that competitors are slowly starting to take, and it's easier to claim now than in 12 months.