Search has a new middleman. Google now answers a growing share of queries itself, with an AI-written summary at the top of the results, and tools like ChatGPT and Perplexity answer millions more questions without sending anyone to a website at all. If your traffic graph has been acting strange lately, this is a big part of why.
The natural reaction is to panic-buy whatever “AI SEO” advice is loudest this week. The better move is calmer: look at what Google actually says, what the data actually shows, and what you can realistically do about it on a WordPress site.
This guide walks through all of it. What AI search changed, the honest traffic numbers (with sources and dates), what genuinely helps your pages get cited, the llms.txt question everyone keeps asking about, and a candid section on what nobody knows yet. Let’s get into it. 👇
What AI search actually changed (and what it didn’t)
Quick vocabulary first. AI Overviews are the AI-written summaries Google shows above the regular results. AI Mode is Google’s fully conversational search tab. ChatGPT search and Perplexity are answer engines: they read the web, then write an answer with a handful of citations. Different products, same direction: the answer increasingly happens before the click.
Here’s the part most hot takes skip. Google’s own documentation (“AI features and your website” on Search Central) says AI Overviews and AI Mode have no additional technical requirements beyond being indexed and eligible to show a snippet. No special schema.org markup. No secret tag. No new file standard. Standard SEO best practices apply: crawlable pages, sensible internal links, good page experience, key content in text form, and structured data that matches what’s actually visible on the page.
You also keep the normal controls. nosnippet, data-nosnippet, and max-snippet limit what Google can quote, noindex keeps a page out entirely, and the Google-Extended token lets you opt out of AI training. Traffic from AI Overviews and AI Mode shows up inside Search Console’s regular “Web” search type, which matters for a reason you’ll see in the uncertainty section below.
Google’s pitch is that clicks coming from pages with AI Overviews are “higher quality,” meaning people stick around longer after arriving. Treat that as the company’s framing rather than an independent finding, and hold it up against the numbers in the next section.
The honest traffic numbers (with sources) 📉
Plenty of studies are floating around, and they don’t all measure the same thing. These are the ones worth knowing, with dates attached, because this space moves fast and last year’s stat can be this year’s folklore.
| Study | What it measured | Headline finding |
|---|---|---|
| Ahrefs (February 2026 update) | 300,000 keywords, December 2023 vs December 2025 Search Console data | Position-1 CTR drops 58% when an AI Overview appears |
| Seer Interactive (through Sept 2025, updated April 2026) | 25.1 million impressions | Organic CTR down 61% on AI Overview queries; cited pages earn more |
| Pew Research (2025) | Real user browsing behavior | 8% of AI-summary visits end in a click vs 15% of visits without one |
| SparkToro/Similarweb (June 2026) | US clickstream data, January to April 2026 | About 68% of searches end with no click at all |
The Ahrefs analysis is the one to sit with. When an AI Overview is present, the top organic result’s click-through rate falls by 58%. Positions 5 and 10 lose 32.6% and 19.4% respectively. In other words, the better you rank, the more an AI Overview costs you. That’s an uncomfortable sentence for anyone who spent years climbing to position one.
Seer Interactive tracked 25.1 million impressions through September 2025 and found organic CTR on queries with AI Overviews fell 61% (from 1.76% to 0.61%), with paid CTR down 68%. Their April 2026 update adds two nuances that matter: the decline has partially leveled off, and pages that get cited inside the overview earn substantially more clicks than pages that don’t. One industry tracker summary puts the citation premium at around 120% more clicks per impression.
Pew Research watched actual browsing behavior in 2025: people clicked a traditional result in only 8% of visits that included an AI summary, versus 15% of visits without one (surfaced in this stats roundup). The citations inside the summaries fared even worse, getting clicked in roughly 1% of visits. People mostly read the summary and leave.
And the baseline underneath all of this: about 68% of US Google searches already ended without any click at all, per SparkToro and Similarweb clickstream data from January to April 2026 (published June 2026, via this state-of-citations report). AI summaries are accelerating a trend that was well underway, not starting one.
Google’s counter-position: AI Mode passed 1 billion monthly users (announced at Google I/O 2026), and the remaining clicks are supposedly better ones. Read the whole picture directionally: fewer clicks overall, and the sites that get cited capture a bigger share of what remains. Being the citation is the new position one.
What actually helps you get cited
No guaranteed tricks exist here (anyone selling one is selling something else). But the documented consensus across Google’s documentation and practitioner studies points to a short, refreshingly unglamorous list:
- Rank well the boring way first. Ahrefs data shows 38% of AI Overview citations come from pages already ranking in the top 10. Crawlability, internal links, page experience, genuinely useful content: the fundamentals do most of the work. AI visibility is mostly SEO wearing a new hat.
- Answer first, then explain. Put the direct answer in the first sentence or two under a clear heading, then go deeper. Question-and-answer formatting is easy for both humans and machines to quote.
- Publish things AI can’t find elsewhere. Original data, real tests, surveys, benchmarks. An AI summary has to cite something. Be the something.
- Use named authors with real expertise. Identifiable, credible authors are consistently part of the citation consensus. “By Staff” inspires no one, human or machine.
- Keep schema accurate and honest. Structured data (the behind-the-scenes code that labels your content for machines) should match what’s visible on the page. Google explicitly calls this out, and mismatches are a trust killer.
- Decide deliberately about AI crawlers. Your robots.txt file controls whether bots like GPTBot (OpenAI), PerplexityBot, and ClaudeBot (Anthropic) can read your site. Blocking them opts you out of being cited by those tools; allowing them is a bet that citations pay off. Either way, make it a choice, not a default you never noticed. 🤖
One more lever: freshness. AI systems favor current information, so keeping your best posts updated matters more than ever. There’s a full Themeisle guide on fighting content decay if you want the deep version, and FreshRank, a free tool from our family, can show you which posts are losing ground so you know where to start.
The llms.txt verdict: harmless, but expect nothing
If you’ve read anything about AI SEO lately, you’ve met llms.txt: a Markdown file proposed in 2024 that sits at your site’s root and maps your key content for large language models. Think of it as a sitemap aimed at chatbots.
The evidence that it does nothing today is about as strong as negative evidence gets. Ahrefs analyzed 137,000 sites (June 2026) and found 97% of llms.txt files never get read by any bot, and 77% of the bots that do read them aren’t AI tools at all. Google doesn’t support it: Gary Illyes said so publicly in July 2025, John Mueller compared it to the meta keywords tag (a ranking signal the industry retired decades ago, which is not a compliment), and Google’s own 2026 AI optimization guidance states llms.txt has no effect on Search rankings or AI features. Adoption sits around 10% of domains (SE Ranking, across roughly 300,000 domains checked), and no major AI provider, whether OpenAI, Anthropic, Google, or Perplexity, has confirmed using it as a discovery or citation signal.
So why does it keep getting recommended? Partly hope, partly vendor marketing. Some vendor blogs claim llms.txt boosts AI visibility, but those claims don’t survive contact with actual crawler logs.
💡 PRO TIP: Adding an llms.txt file costs almost nothing, and both Yoast and Rank Math can generate one for you. Go ahead if it makes you happy. Just set your expectations to zero, and spend the saved energy on the citation fundamentals above.
AI search optimization on WordPress: the practical setup
The good news for WordPress site owners: most of this machinery is plugin territory, and the major SEO plugins have already shipped their AI-era features. Here’s what does what:
- Schema, handled (Free). Both Yoast SEO and Rank Math output structured data automatically, even on their free tiers. Your job is the honest part: make sure the markup matches what’s visible on the page, and fill in author names properly instead of leaving them blank.
- AI-assisted metadata (Pro/Premium). Rank Math’s Content AI (Pro) bundles 40+ AI tools, and Yoast’s AI Generate and AI Optimize features (Premium) draft titles and meta descriptions for you. Handy for throughput, though neither one directly makes AI search cite you.
- Brand monitoring in AI answers. Yoast’s AI Brand Insights (part of its AI+ bundle) tracks how your brand shows up in ChatGPT and Perplexity answers. It’s one of the few ways to measure the citation side of this instead of guessing.
- llms.txt generation. Both plugins will generate the file for you. See the verdict above before getting excited.
One thing not to conflate: WordPress.com’s built-in AI Assistant (launched February 2026) helps you edit your site, adjust styles, and create images. It’s a site-management assistant, not an AI search optimization feature. Useful tool, different job.
What we don’t know yet 🤔
Most articles on this topic pretend to certainty. Here’s an honest list of open questions instead:
- Whether the citation premium holds. Seer’s data shows cited pages earning more clicks, but AI Mode is young and user behavior around it is still forming.
- How much traffic AI features take from your specific site. Search Console lumps AI Overviews and AI Mode traffic into the general “Web” bucket, so there’s no clean per-site number today.
- What AI providers actually optimize for. None of them publish citation criteria. Every “best practice” is inferred from observed patterns, not official documentation.
- How ChatGPT and Perplexity referrals behave at scale. Most solid CTR studies measure Google’s AI features; public data on the standalone answer engines is much thinner.
- Whether llms.txt (or something like it) ever gets adopted. Possible, but today’s evidence says don’t wait up. 😅
So treat every statistic in this space as directional, not gospel. The studies disagree on magnitudes because they measure different things. Anyone claiming certainty right now is either overconfident or has a course to sell.
The bottom line
- AI Overviews and answer engines mean fewer clicks overall. The drop is real, sourced, and worth planning around.
- Getting cited is the new front page: fundamentals first, answer-first structure, original data, named authors, honest schema.
- llms.txt is harmless, optional, and currently supported by zero evidence. File under “fine, whatever.”
- WordPress tooling (Yoast, Rank Math) already covers schema and most of the extras.
The sites that win the AI search era will look a lot like the sites that won the previous one: genuinely useful, clearly written, technically sound. The main difference is that being quotable now matters as much as being clickable.
Have AI Overviews started citing your content, or eating your traffic? Let us know in the comments section below!
FREE GUIDE
4 Essential Steps to Speed Up Your WordPress Website
Follow the simple steps in our 4-part mini series and reduce your loading times by 50-80%. 🚀
