Google launched AI Mode as an experimental search feature on March 5, 2025. Eighteen months later, AI Overviews sit above the organic results on roughly 30% of Google searches, ChatGPT has run its own web search product since October 31, 2024, and a term that did not exist before November 2023, Generative Engine Optimization (GEO), now has its own line item in agency pricing sheets. That is the actual timeline behind "AI is changing SEO." Not a vibe, a sequence of shipped product launches with dates attached.
This piece works through what changed, what the data says about the damage (and the parts that are not damage at all), and what using AI in SEO actually looks like for a team heading into the rest of 2026, as distinct from just using AI for SEO grunt work like drafting and research.
What does AI in SEO actually mean, and what changed between 2024 and 2026?
Three products did most of the work. OpenAI shipped ChatGPT search on October 31, 2024, giving a chatbot with hundreds of millions of weekly users a built-in way to cite live web pages instead of just its training data. Google answered by upgrading AI Overviews to Gemini 2.0 and introducing AI Mode as an experimental tab on March 5, 2025, according to Google's own announcement. By late 2024, Google said AI Overviews were already used by more than a billion people a month, before AI Mode even existed.
Perplexity had been running an answer-engine model since 2022, but 2024 to 2026 is when it stopped being a niche tool for early adopters and started showing up in marketing dashboards as a real referral source. The shared mechanic across all three products is the same: an LLM reads several web pages, writes an answer in its own words, and the user often never clicks through to any of them.
That mechanic is what "AI in SEO" actually means in practice: an extra layer of AI systems that read, summarize, and sometimes replace the click a page used to get from ranking well. Ranking is not enough anymore; the page has to survive being paraphrased.
Is AI search actually taking clicks away from real websites?
Yes, on a specific slice of queries, and no, not evenly across the board. Graphite's analysis of Similarweb data across more than 40,000 of the largest US websites found organic traffic down 2.5% year over year, comparing February to December 2024 against January to November 2025, as reported by Search Engine Land. That is a real decline, and it is nowhere close to the 25% drop Gartner projected in its February 2024 press release forecasting that traditional search engine volume would fall by 2026.
The 2.5% figure hides a much sharper effect where it actually matters. On the roughly 30% of queries where an AI Overview appears, Graphite's own data shows organic click-through rate down 35%. Seer Interactive's campaign-level study of 42 client organizations and 3,119 search terms found something steeper on that same kind of query: organic CTR fell from 1.76% to 0.61% between June 2024 and September 2025, a 61% decline over 15 months.
The pattern is consistent: pages that answer a broad, definitional query lose clicks to the summary sitting above them. Pages that serve a transactional, branded, or highly specific query keep most of their traffic, because the AI answer cannot substitute for actually visiting the site. For a full breakdown of where these numbers actually come from, including the two different studies both cited as "the 61% figure," see our separate piece on whether SEO is dead in 2026.
What is Generative Engine Optimization, and how is it different from SEO?
GEO is the practice of getting a website's content selected and cited inside an AI-generated answer, rather than just ranked in a list of blue links. The term comes from a specific paper, "GEO: Generative Engine Optimization," published by researchers including Aggarwal and Murahari in November 2023, which formalized the idea that generative engines "satisfy queries by synthesizing information from multiple sources and summarizing them using LLMs," and that content creators have "little to no control over when and how their content is displayed" inside that summary.
Traditional SEO optimizes for a ranking position: get the page into the top 10, then the top 3. GEO optimizes for something else entirely: get a specific sentence, stat, or definition selected by the model as the source of its answer, whether or not that produces a click. Those two goals overlap heavily (both still want crawlable, fast, well-structured pages) but they are not the same objective, and content built purely for the old objective often loses the new one.
How do AI models decide what to cite?
They pull the most extractable, sourced, unambiguous statement available on the topic, not necessarily the highest-ranking page. Three things reliably increase citation odds:
- Answer-first structure. If a heading poses a question, the sentence directly under it needs to answer that question in full, because models lift the first sentence rather than reading the whole section for context.
- Structured data. FAQPage and Article schema give a model a machine-readable shortcut to the same content a human would have to parse from prose, and Google's own developer documentation on AI features points crawlers toward exactly this kind of structured markup.
- A plain-text map of the site. The llms.txt convention, first proposed in 2024 and now on its second published version, gives AI crawlers a concise list of a site's key pages instead of forcing them to parse full HTML navigation and JavaScript.
None of that is exotic. It is the same discipline good technical SEO always rewarded: clear structure, real markup, no ambiguity about what a page is actually claiming. GEO did not invent a new skill, it raised the cost of skipping the old one.
Is Google penalizing AI-generated content in 2026?
No, not for being AI-generated specifically. Google folded its Helpful Content System into its core ranking systems in March 2024, and its public guidance since then has been consistent: the target is low-quality, unhelpful content produced at scale to game rankings, regardless of whether a human or a model wrote it. Content written by AI that is accurate, well-sourced, and genuinely useful is not automatically penalized; content that is thin, unverified, or written to match a keyword rather than answer a question performs badly whether a person or a model typed it.
That distinction matters because a lot of 2026 anxiety about "AI content penalties" is really anxiety about the same low-quality-content problem search engines have targeted for over a decade, just with a new production method attached to it. The actual risk in 2026 content is unverified claims, not authorship. Publishing a repeated statistic without checking which study it actually came from is the kind of error that damages trust with both readers and AI systems that are increasingly cross-checking claims against multiple sources before citing them.
Traditional SEO vs. GEO: what actually differs
| Dimension | Traditional SEO | GEO / AEO |
|---|---|---|
| Primary goal | Rank in top 10 organic positions | Get selected as the cited source in an AI answer |
| Success metric | Rankings, organic clicks | Citation frequency, brand mentions inside AI answers |
| Content shape rewarded | Long-form pages targeting a keyword | Answer-first sections, extractable facts, clear definitions |
| Technical layer | Core Web Vitals, crawlability, backlinks | Structured data (FAQPage, Article schema), llms.txt, entity clarity |
| What breaks it | Thin content, weak links, slow pages | Vague claims, buried answers, no schema, unsourced stats |
What does using AI for SEO actually look like in 2026?
Start by treating GEO as an addition to existing SEO work, not a replacement for it. A technical SEO audit still catches the crawlability and speed problems that block AI crawlers just as much as Googlebot. From there, three changes matter most:
- Rewrite your FAQ and definition sections answer-first. The first sentence under any question heading should be the complete, factual answer. Everything after it can add nuance.
- Ship real structured data, not decorative markup. FAQPage and Article schema should match the visible content exactly, since Google treats a mismatch between markup and page content as a manual-action risk, not a shortcut.
- Publish (and maintain) an llms.txt file. It costs little to produce and gives AI crawlers a direct map to your highest-value pages instead of making them guess from a sitemap built for humans.
Our GEO optimization service builds exactly this stack for clients, and our AI SEO service covers the crawlability and entity work underneath it. For teams producing a lot of content, citation-optimized content work matters more than raw volume: one well-sourced, answer-first page tends to outperform five generic ones in actual citation rate.
None of this replaces link building, page speed, or writing for real search intent. It sits on top of that foundation. A site with weak fundamentals will not out-cite a competitor with strong fundamentals just by adding schema.
How do you know if AI is already affecting your specific traffic?
Check Google Search Console for queries where impressions are flat or rising but clicks are falling; that pattern is the fingerprint of an AI Overview absorbing clicks on a query you still technically rank for. Then check server logs or a tool like our AI visibility audit for whether GPTBot, PerplexityBot, and Google-Extended are actually crawling your key pages, since a site that blocks those crawlers in robots.txt cannot be cited even if the content would otherwise qualify.
If the pages losing clicks are broad, definitional, top-of-funnel content, that is expected and matches the industry-wide 30-35% CTR pattern described above. If transactional or branded pages are losing clicks, that is a different, more urgent problem worth investigating separately, since AI answers rarely substitute for those query types.
Frequently asked questions
What does "AI in SEO" actually mean in 2026?
It means two separate things that get conflated: using AI tools to do SEO work faster (research, drafting, technical audits), and optimizing content so AI-powered search products like ChatGPT search, Google AI Overviews, and Perplexity choose to cite it. The second one, often called GEO or AEO, is the part actually reshaping strategy, since it changes what "ranking well" is even for.
Is GEO replacing SEO?
No. Using AI for SEO tasks like drafting and technical audits is not the same thing as GEO, which adds a citation-focused layer on top of standard SEO fundamentals like crawlability, page speed, and backlinks; it does not replace them. A page with weak technical SEO will not get cited by an AI model even with perfect schema, because the model still has to be able to crawl and parse it first.
Does Google penalize content written by AI?
No, not for authorship alone. Google's Helpful Content System, folded into core ranking in March 2024, targets low-quality and unhelpful content regardless of whether it was written by a person or a model; accurate, well-sourced AI-assisted content is not automatically demoted.
How much organic traffic has AI search actually cost websites?
Across more than 40,000 large US sites, Graphite's analysis of Similarweb data found organic traffic down 2.5% year over year, far below Gartner's 2024 prediction of a 25% decline by 2026. On the specific slice of queries where an AI Overview appears (about 30% of searches), the click-through rate hit is much sharper: 35% per Graphite, and 61% over 15 months per Seer Interactive's campaign-level data.
What is llms.txt and do I need one?
llms.txt is a plain-text file, first proposed in 2024, that lists a site's key pages and one-line descriptions so AI crawlers can map the site without parsing full HTML navigation. It is not required for citation, but it lowers the friction for AI crawlers evaluating a site with many pages, similar to how a clean sitemap helps traditional search crawlers.
Which pages are most exposed to losing traffic to AI Overviews?
Broad, definitional, top-of-funnel informational pages are the most exposed, since an AI Overview can often fully answer that kind of query without a click. Transactional pages, branded searches, and highly specific long-tail queries are far less affected, because completing an action or verifying a specific detail still requires visiting the actual site.
Should a small business still invest in SEO given AI search growth?
Yes. Google still holds over 90% of the search market as of 2026, and most AI-assisted queries supplement rather than replace traditional search for most users. The practical move is auditing which specific pages are exposed to AI Overview cannibalization and prioritizing GEO work there, not abandoning SEO investment across the board.
Sources: Gartner, "Gartner Predicts Search Engine Volume Will Drop 25% by 2026", Graphite, "Debunking the myth that search is dying", Search Engine Land, "Organic search traffic is down 2.5% YoY, new data shows", Seer Interactive, "AIO Impact on Google CTR: September 2025 Update", Google, "Expanding AI Overviews and introducing AI Mode", OpenAI, "Introducing ChatGPT search", Aggarwal et al., "GEO: Generative Engine Optimization", llmstxt.org
This post is part of our GEO Optimization guide. Related reading: 1 in 10 AI Citations Are From Self-Promo, ChatGPT vs Gemini, AI SEO Strategies That Actually Work.
Oussama leads technical and on-page SEO at AY Rank. He specializes in structured data engineering, crawl optimization, and building the entity architecture that makes AI models cite our clients.
Full Bio →
