AI SEO: how to optimize your site for ChatGPT and Google AI Overviews in 2026

In March 2024, Google rolled out AI Overviews to over 100 countries. A year later, ChatGPT Search crossed 400 million weekly active users — and for many people, it became the first reflex for searching, replacing Google. For businesses in Moldova and the wider region, that translates to a hard truth: traffic from classic search results is melting away, and the old SEO playbook no longer works on its own.

AI SEO (also called Generative Engine Optimization, or GEO) is the response to this shift. It isn't a rebranded SEO trend — it's a separate layer that determines whether your brand gets cited in AI-generated answers or ignored entirely. In this guide I cover what AI SEO actually means in 2026, how generative engines work under the hood, and what you need to change on your site this week.

What AI SEO is and why it matters in 2026

AI SEO is the practice of optimizing your content, technical structure, and brand authority so that generative engines — Google AI Overviews, ChatGPT Search, Perplexity, Claude, Gemini — cite you as a source in their answers. Classic SEO chased a position in a list of ten blue links. AI SEO chases a mention inside a paragraph written by a language model.

The difference isn't cosmetic. According to Statista data published in 2025, more than 58% of users who see an AI Overview don't click on any classic organic result. The direct answer is enough for them. If your brand isn't among the cited sources, that traffic simply doesn't exist anymore.

Across projects we audited at XCORE during 2025, pages that hadn't been adjusted for AI search lost between 18% and 41% of organic traffic over 12 months — even when their classic rankings stayed the same. The pages that had been optimized gained on new fronts: brand citations in AI Overviews, referral traffic from Perplexity, conversions from ChatGPT Search.

What it means to be "cited" by an AI

When someone asks "what's the best CRM for a small business in Moldova", the AI doesn't open your site like a regular user. The crawler already read the page, broke it into chunks, indexed it semantically and stored it. At query time, the model retrieves the relevant pieces, synthesizes them into an answer, and — if the quality is high enough — surfaces your link as a source.

Being "cited" means two things: your brand appears as a link in the visible source list under the answer, and concrete fragments of your text get paraphrased or quoted directly inside the answer body. The first brings traffic. The second builds brand authority — even without a click.

How AI Overviews, ChatGPT Search and Perplexity actually work

The three dominant generative engines have different architectures, but they share one foundation: retrieval-augmented generation (RAG). The model retrieves relevant sources from an index or the live web, then generates the answer using them as context.

Google AI Overviews

Uses the standard Google index plus an extra selection layer. The algorithm prefers pages with clean structure (H2/H3, lists, short definitions), correct schema markup, and strong E-E-A-T signals (Experience, Expertise, Authoritativeness, Trustworthiness). It cites 3–7 sources per answer on average.

ChatGPT Search

Runs its own crawler (OpenAI SearchGPT bot) plus a Bing partnership. It rewards conversational content — pages written in natural language, with rephrased questions inside the body, get prioritized. Citations are fewer (1–4 per answer) but more visible.

Perplexity

The most aggressive in citations — 5 to 12 sources per answer, with reference numbers placed directly in the text. Its crawler (PerplexityBot) actively visits smaller sites, which makes Perplexity an accessible channel for local businesses that haven't broken through in Google yet.

AI SEO vs classic SEO: what really changes

A lot of people assume AI SEO is just SEO under a new name. That's an expensive assumption. Here's where the two diverge fundamentally:

• Target — classic SEO optimized for a keyword. AI SEO optimizes for a conversationally phrased question. "Crm moldova" becomes "what's the best crm for a small business in Chisinau".
• Content format — classic SEO rewarded long, dense articles with optimized keyword density. AI SEO rewards chunks — short, self-contained paragraphs that answer one specific question and can be lifted as-is.
• Authority signals — backlinks still matter, but unlinked brand mentions, reviews, citations in industry articles, and presence on trusted platforms weigh much more in AI search than they did in classic SEO.
• Structured data — schema markup is no longer optional. FAQPage, Article, Organization, Product, HowTo — these are signals the AI uses directly to decide whether your content is "extractable".
• Velocity — a fresh article can show up in AI Overviews within 24–72 hours. The same article needs 4–8 weeks to stabilize in classic SERP. AI search rewards freshness.

What doesn't change: content quality. A shallow text won't be cited in AI search or in Google. The difference is that AI SEO is far less forgiving with filler — a language model spots an empty paragraph instantly and skips it.

How to optimize content to be cited by AI

On to the practical part. This is what we implemented on XCORE client sites in 2025 that drove growth in AI citations.

1. Write for extraction, not for reading flow

Each H2 should answer a single question. The first paragraph under each H2 should contain the direct answer in 2–4 sentences, before any context or examples. The AI scans this paragraph and decides whether it's worth extracting. If you start with "before we talk about X, let's discuss the context…", the paragraph is ignored.

2. Use explicit definitions

"X is a Y that does Z" sentences are gold for AI. Structured, short, unambiguous. Open articles with an explicit definition of the main concept — exactly like in this guide, paragraph 4.

3. Implement schema markup correctly

For blog articles: Article + FAQPage (when there's a Q&A section). For service pages: Service + Organization. For tutorials: HowTo. Schema with mismatched data is worse than no schema — AI catches inconsistencies and penalizes the page.

4. Build off-site brand authority

Mentions on local news outlets (for the Moldovan market: point.md, agora.md, mold-street.com), in business podcasts, in industry articles — they all feed the authority signals AI uses to pick its sources. One well-placed mention in local press outweighs twenty generic backlinks.

5. Add concrete E-E-A-T signals

Project counts, years of experience, real cases, measurable numbers. "We've been working for 8 years with over 120 clients in Moldova" is a much stronger signal than "we have extensive experience in the field". AI recognizes specificity and rewards it.

6. Don't skip llms.txt

Proposed in 2024 as a standard, the llms.txt file at your site root gives AI a clean summary of your most important content. It's supported by Anthropic, Perplexity and parts of the AI ecosystem. One hour to implement. The benefit — direct control over how AI perceives your site.

Practical AI SEO checklist for 2026

If you have 30 minutes this week and want one concrete first step, run the list below on your most important page:

1. Check AI indexing — search your brand name in ChatGPT and Perplexity. If you don't show up, you have a citation problem, not a ranking problem.
2. Audit the opening paragraphs of every key page — do they answer the main question directly, or do they start with throat-clearing?
3. Add FAQPage schema on at least 3 trafficked pages — the impact in AI Overviews shows up in 2–4 weeks.
4. Add llms.txt — 30 minutes of work, opens a new channel.
5. Look for PerplexityBot and OAI-SearchBot in your logs — if they're missing, your robots.txt is blocking them. Verify and explicitly allow them.
6. Build 1–2 off-site mentions per month — opinion pieces, interviews, expert commentary. Links are a bonus, the mention itself is the main asset.
7. Rephrase 3 H2s in each article into a question — start with your highest-converting page.

If you want to discuss a full audit for your business, the XCORE team offers AI SEO services for the Moldovan and broader markets — we combine classic technical analysis with an AI citation audit and a 90-day execution plan.

Mistakes that crush your visibility in AI search

The list below is built from the errors we kept seeing in audits over the past year. Each one costs real money — either in lost traffic or in missing citations.

• Unedited AI-generated content — the paradox: AI recognizes text written by another AI and trusts it less. Raw ChatGPT output without serious human editing gets cited less than content from real authors.
• Long paragraphs without structure — an 800-word block with no H3, lists or breaks. The AI can't extract chunks from that and moves on.
• Incorrect schema markup — data that doesn't match the visible content. Google and AI engines compare the two; any mismatch gets penalized.
• Missing author — articles without a byline, bio and link to the author profile send a weak E-E-A-T signal. AI prefers sources with clear identity.
• Overly aggressive robots.txt — preemptively blocking all AI crawlers takes you out of the ecosystem entirely. Decide explicitly which ones to let in.
• Ignoring language variants — for the local market, content in RO and RU gets cited in AI answers generated in those same languages. An English-only site loses massive local traffic.

Frequently asked questions

Bottom line: start now, not in six months

The shift from classic SEO to AI SEO isn't a slow transition — it's a fast redistribution of user attention. Brands that start optimizing in 2026 catch the wave. Those that wait for "things to settle down" will find a year from now that 3–5 dominant voices already own every AI answer in their niche, and the cost of breaking into that list has grown tenfold.

If you want a concrete starting point, begin with an AI SEO audit on your highest-traffic page. The XCORE team can fold that analysis into your classic SEO and AI integration strategy in a single plan — because in 2026, those three aren't separate disciplines anymore, they're the same one.