Two weeks ago one of our B2B clients showed me their Search Console data. Keyword ranking: unchanged for 11 months at position #2. Clicks: down 47% year over year. They'd spent the previous quarter blaming the content team, the developers, a recent redesign. None of it was the cause.
Google AI Overviews rolled out globally through 2024–2025. They now show on 40%+ of queries in English and 15–25% in smaller languages (we see Polish at the lower end of that range in 2026). For the same #1 ranking position, CTR drops from ~40% without an AI Overview to ~18–22% with one above you. Up to 60% less traffic at exactly the same ranking.
SEO didn't die. CTR did. The good news: pages cited INSIDE the AI Overview often get 30–50% CTR — sometimes higher than pre-AI traffic at the same position. The game stopped being "rank #1". The game became "be the source AI trusts".
The cause was a small grey box that now sits above their result in Google SERP, labelled "AI Overview". Their page is still ranked #2. It just isn't being clicked anymore, because the AI above it already answered the question and cited somebody else.
I have seen this exact pattern in 8 of the 11 SEO audits I ran in Q1 2026. It is the single biggest shift in organic search since the mobile-first index in 2018, and most SEO teams I talk to are still one conversation behind it.
This article is not a generic "GEO 101" post. It is what I actually do when a client asks me to prepare their site for AI search in 2026: the exact tactics that moved the needle in our own tests, the ones that turned out to be marketing noise, and the measurement stack I run with every client. Aimed at heads of SEO, content leads and founders whose organic traffic stopped growing even though their rankings didn't.
What is GEO (and what it is not)
Generative Engine Optimization (GEO) is the practice of optimising content so it gets cited, summarised and linked correctly by AI answer engines. Instead of optimising only for ranking in a list of blue links, you optimise for being the source that the AI extracts from when it writes its answer.
Key AI engines to optimise for in 2026:
- Google AI Overviews — the AI answer above Google's organic results.
- ChatGPT Search — OpenAI's search product with citations.
- Perplexity — citation-first AI search.
- Bing Copilot — Microsoft's AI search.
- Claude web search — Anthropic's Claude with search enabled.
- Gemini with Google Search grounding.
GEO is what SEO was in 2005 — a new discipline where best practices are still forming. The companies that get this right for two quarters have an advantage that will be hard to catch.
What I'm actually seeing in client analytics
Four patterns that have appeared in every recent audit I have run. These are not industry headlines — they are what I see when I log into a mid-market B2B company's GA4 and Search Console.
1. The "same rank, fewer clicks" pattern
This is the single most common finding. Keywords holding their ranking, impressions flat or up, clicks collapsing. In the last 8 audits we ran, the median CTR drop on informational queries with an AI Overview was 41%. For pages at position #1, the drop is steeper — in one manufacturing client's case, their flagship pillar page lost 54% of its clicks while holding position #1 for 6 months. No algorithm penalty. Just a new grey box above them.
Source: aggregate of Ahrefs 2025 study + 8 B2B client audits run by Hauer Power in Q1 2026 (N=80+ queries). Note the green bar: pages actually CITED inside the Overview outperform pre-AI CTR — the game is citation, not displacement.
2. ChatGPT referrals becoming meaningful
In January 2025 I told a client they should stop worrying about ChatGPT referrals because the volume was "statistical noise". I was wrong. By October 2025 that same client was getting 6.2% of their organic-equivalent sessions from ChatGPT citations. By March 2026: 11%. Every SaaS and consulting client I work with now has chat.openai.com and perplexity.ai in the top-20 referrers list. Five years ago this would have been considered a massive channel discovery.
3. B2B research is shifting upstream into AI
Before: buyer Googles "best CRM for manufacturing" → lands on comparison article → shortlists vendors. Now: buyer asks ChatGPT → gets 3-4 named vendors in the answer → goes directly to those vendors' sites. If you are not in the AI answer, you never enter the consideration set. I watched this happen live in a user research session in February — the buyer never touched Google for the shortlisting step. Similarweb data from late 2025 confirms this is systemic, not anecdotal.
4. Brand mentions in AI answers are a new pipeline input
For one of our SaaS clients, we tracked what happened to brand search volume after we made their site GEO-friendly. Over 5 months, branded searches on Google rose 34% — and roughly half of the incremental brand search users had first seen the brand mentioned in an AI answer to a non-branded query. Being cited in AI answers compounds into brand equity that eventually shows up in classic organic search too.
How AI engines actually pick sources (what I reverse-engineered)
I ran a small internal study in March 2026: took 40 informational B2B queries in our niche, ran them through Google AI Overviews, ChatGPT Search and Perplexity, logged every cited source, and cross-referenced with Ahrefs for ranking position. Here is what the data actually says — not what vendors promise.
1. Traditional ranking is the gate (and it is harder than you think)
Across 40 queries × 3 engines = 120 AI answers, 94% of cited sources were ranked in Google's top 20 organic for the same query. The remaining 6% were almost all brand pages (the AI knew the brand was relevant even if the specific page was not top-ranked). The practical implication: you cannot "GEO your way" into AI answers without first ranking organically. GEO is the layer on top, not the replacement.
2. Fact density beats word count
When I clustered the cited pages by structure, the winning pages averaged 1 specific factual claim (number, stat, named entity) per 120 words of body text. Losing pages at the same rank averaged 1 per 400 words. Fact density matters more than page length. "Response time under 5 minutes increases B2B conversion by 21x" is exactly the kind of sentence engines lift. "Fast response times are important" is not.
3. Entity and topic clarity (schema actually helps here)
Pages with complete Article + FAQPage + BreadcrumbList + Organization schema were 2.3x more likely to be cited than structurally equivalent pages without schema, in my sample. This is not proof of causation, but the correlation is strong enough that I now treat schema as non-negotiable for any page I want AI engines to quote.
4. E-E-A-T signals, reweighted
Named author, date, sameAs to LinkedIn, outbound links to primary sources — all correlated with citation frequency. The strongest signal in my data: pages with a first-person author statement ("I've deployed this at X clients…") were over-represented among cited sources. LLMs appear to preferentially extract text that reads as expert testimony.
5. Recency matters more on some queries than others
For evergreen queries ("what is a CRM") recency barely mattered. For time-sensitive queries ("best AI tools 2026", "AI Act deadline") recency was brutal: pages older than 6 months were cited in <20% of answers even when ranking in top 5. Keep dateModified current and visible in the body.
The Princeton et al. academic paper on GEO from late 2023 tested similar tactics in a controlled setting. Their top three (citing sources with quotation, adding statistics, using structure) match what I saw in the wild two years later. The research held up.
10 tactics I actually deploy (with the ones that failed)
I have tested these on our own site (hauerpower.com) and on 6 client sites over the last 4 months. Tactics marked ✅ produced measurable citation lift within 6-10 weeks. The ones marked ⚠ are industry advice that did not move the needle in my tests.
1. Rank first in traditional SEO
Non-negotiable. In our 40-query study, 94% of cited sources were already top-20 organically. If your classical SEO is broken, fix it before touching GEO. Start with our SEO audit guide.
2. Answer the question in the first 100 words
I tested this on 12 existing blog posts. Rewrote the opening to be a direct one-paragraph answer with the key claim bolded. Six of the twelve started getting cited in AI Overviews within 8 weeks. The other six were below organic position 10 — GEO cannot overcome bad ranking.
3. Specific numbers over round-ups
"Customer acquisition cost is going up" → zero citations. "B2B SaaS CAC averaged $702 in 2025 per Incredo 2025 State of SaaS report" → cited in 4 AI answers within 6 weeks. Sprinkle specific, sourced numbers throughout. Every section should have at least one.
4. Outbound links to primary sources
Counter-intuitive move that many SEOs refuse to make: link out to .edu, .gov, original research, vendor docs. I added 2-3 outbound links per post across our top 15 articles. Citation rate in AI answers roughly doubled. AI engines verify facts against what you cite — make it easy for them. We do this on our B2B marketing article and the numbers followed.
5. Structure for extraction (aggressive H3 use + comparison tables)
Long walls of text get excerpted poorly. I refactored 4 of our longest articles into 800-1200 word sections, each with its own H2/H3 and 2-4 short paragraphs. Average citation rate doubled. Bonus: dwell time went up because the articles became easier to skim.
Inside those sections, comparison tables are disproportionately cited. When AI engines answer "X vs Y" type queries, they lift rows directly out of HTML tables. On our own comparison posts (DeepSeek vs ChatGPT, ChatGPT vs Gemini), the tables are the single most-cited element according to our server logs. If your topic has any comparative angle, ship a table.
6. FAQPage schema — biggest single-change win
If you only do one thing from this list, do FAQPage JSON-LD. We added it to 23 of our pages in February. AI Overview citations tracked via Peec AI rose from 14 to 51 over 8 weeks. It is the cheapest intervention with the largest effect I have measured.
7. Original research and benchmarks
Want to be THE source AI cites on a topic? Publish original numbers nobody else has. Small internal studies (N=40 is fine if documented), surveys of your customer base, benchmarks from your operations. One 400-word original data piece I published in February has been cited by AI answers 200+ times according to our referral logs. That is not replicable with curated content.
8. First-person expert voice
"Studies show…" is generic. "I have run 11 SEO audits in Q1 2026, and in 8 of them the pattern was…" is quotable. I rewrote bylines across our top content to include first-person claims with specific numbers from our own experience. The over-representation of first-person text in AI citations is consistent enough that I treat it as a ranking factor.
9. dateModified + visible "Last updated"
For time-sensitive topics, refresh quarterly and update both the schema and a visible line in the body. Our quarterly refresh cycle bumped citation frequency on "best AI tools 2026"-type queries by about 60% compared to pages we left alone.
10. llms.txt at domain root — interesting but unclear lift
I added llms.txt to hauerpower.com in April with our 30 best pieces curated. In server logs I see OpenAI, Anthropic and Perplexity bots fetching it. But I have not yet seen a clear correlation between llms.txt presence and citation lift in our small sample. Do it — it is 10 minutes of work — but do not expect magic. My guess is it becomes more important as the ecosystem matures.
Things that did NOT work in my tests
- Pure keyword density: no effect. I tested dense vs natural-language versions of the same content. AI engines picked the natural version.
- Brute forcing article length: I padded a 1500-word post to 4000 words. Citations went down, not up. Length without substance = worse, not better.
- "AI-friendly" writing patterns pushed on LinkedIn (e.g., "write answers that start with 'Yes, …'"): negligible effect in my tests. Write for humans first.
- Paying for PR just to hit AI answers: one client tried this. AI engines are surprisingly good at detecting promotional tone. No citation lift.
Technical setup for GEO
Required foundations
- Clean robots.txt with AI user agents explicitly allowed (or not blocked).
- Sitemap.xml with lastmod dates.
- Canonical tags on every page.
- Article schema with author, datePublished, dateModified.
- FAQPage schema where applicable.
- BreadcrumbList schema.
- Fast Core Web Vitals (LCP, INP, CLS in "good").
GEO-specific additions
- llms.txt at domain root — curated index for AI crawlers.
- Author schema with sameAs linking to LinkedIn and professional profiles.
- Organization schema on homepage with sameAs to social, publisher pages.
- Person schema for author pages.
- Specific dates visible in article body, not just schema.
- HTTPS, no mixed content — AI engines deprioritise insecure sources.
The exact measurement stack I run
GEO measurement is where most teams stop. There is no single "rank tracker for ChatGPT" that gives you Search Console-level clarity yet. Here is the stack I actually use on client retainers — not theoretical, operational.
Weekly (automated)
- Peec AI — tracks citations in ChatGPT, Claude, Perplexity and Google AI Overviews for a set of target queries. We track 80-120 queries per client. $50-150/month, worth it.
- Server-log dashboard — I parse access logs weekly for AI user agents (OAI-SearchBot, ClaudeBot, PerplexityBot, GPTBot, Google-Extended, anthropic-ai). Rising fetch frequency predicts rising citations 2-4 weeks later. Simple Python script writes to Google Sheets.
- GA4 referral reports — filtered view for chat.openai.com, perplexity.ai, bing.com/chat, you.com, phind.com. New traffic channel that did not exist in 2023.
Monthly (manual — cannot fully automate this yet)
- Manual SERP sampling: 20 target queries across ChatGPT, Claude, Perplexity and Google. Is the brand mentioned? Cited? Correctly? Documented in a spreadsheet with screenshots.
- Search Console CTR regression: for each target query, compare current CTR to same position 12 months ago. Drops >30% with stable position = AI Overview pressure.
- Brand search Trend: Google Trends for branded searches. If GEO is working, branded search rises over time even if direct clicks do not.
What I stopped using
Free AI rank trackers that scrape ChatGPT without an API contract. Accuracy was inconsistent and results shifted based on session fingerprint. Paid tools with proper API relationships or dedicated browser automation give much cleaner data.
Myths and bad advice
- "Keyword density still matters." No more than in classic SEO. Write for humans.
- "Long content always wins in AI answers." Wrong. Clear, structured content wins. Bloating to 5000 words doesn't help.
- "Put GEO-specific prompts in your content to manipulate AI." Don't. Prompt injection in public content violates terms and gets detected.
- "Block AI crawlers to protect content." Rarely the right call for B2B marketing content. You're invisible to AI search = you lose the traffic you'd otherwise get.
- "Replace SEO team with GEO team." You need one team doing both. GEO without SEO foundations doesn't work.
For practical SEO foundations, see our guides on SEO audit, SEO content and meta tags.
FAQ
What is Generative Engine Optimization (GEO)?
Generative Engine Optimization (GEO) is the practice of optimising content so it gets cited and summarised correctly by AI answer engines like ChatGPT, Claude, Perplexity, Google AI Overviews and Bing Copilot. It extends classic SEO (ranking in a list of links) into a new dimension: being the source the AI quotes from, not just one of ten blue links.
Is GEO different from SEO?
Overlapping, not identical. Both care about authority, structured content and crawlability. SEO optimises for click-through on a search result link. GEO optimises for being cited inside an AI answer, where the user may never click through. Classic SEO signals (domain authority, quality backlinks, clean technical SEO) remain foundational; GEO adds: structured fact-dense content, clear topical entities, llms.txt, explicit source attribution.
How do AI engines pick sources?
Four signals dominate in 2026: (1) traditional SEO ranking, top 10 organic results are the candidate set for most AI engines; (2) citation-worthiness, clear facts, data, quotes that are easy to excerpt; (3) entity clarity, content that makes the subject unambiguously identifiable; (4) E-E-A-T signals, author bylines with credentials, dates, sources. AI engines are essentially doing extractive summarisation over high-ranking sources.
Can I stop doing traditional SEO and only focus on GEO?
No. In 2026, AI answer engines still pull overwhelmingly from content that ranks well in classic search. A page that doesn't rank in Google's top 20 is essentially invisible to Google AI Overviews, and mostly invisible to ChatGPT search. GEO is SEO + new layer, not SEO replacement. Do both.