Over the last 18 months, almost every conversation we have with clients eventually turns to AI. One of the most frequent questions clients have is how AI is impacting their site traffic.

We wrote a post last November on Google’s AI Overviews and what they mean for your site. A lot has changed since then. AI Overviews are more common, tools like ChatGPT and Claude now have hundreds of millions of weekly users, and the data on how all of this affects websites has become clearer.

This post is a follow-up FAQ for anyone trying to make sense of the landscape in 2026.

How AI is affecting website traffic

The short answer: AI is impacting your site traffic a lot, and more every month.

A few numbers worth knowing:

• AI Overviews now appear in roughly 25% of all Google searches in the US, up from about 13% in early 2025, according to Conductor’s 2026 benchmarks.
• When an AI Overview appears, click-through rates drop. A Pew Research Center study found CTR falls from 15% to 8%.
• Only about 1% of users click links inside AI Overviews.
• Non-branded informational traffic is down 15–30% year-over-year, according to Search Engine Land.

There is also a second trend worth watching. People are increasingly skipping Google entirely and going straight to ChatGPT, Claude, Gemini, or Perplexity. A March 2026 analysis from Graphite found that monthly sessions on AI tools are now the equivalent of roughly 56% of global search volume. ChatGPT alone processes an estimated 2.5 billion prompts per day.

Does informational content still matter

Informational pages (explainer content, how-tos, definitions, guides) are seeing the biggest traffic declines. On the surface, that sounds like pure bad news. But informational content is also the content most likely to power AI responses, and that can be hugely valuable.

Here is the shift to understand. A page that used to get 10,000 visits a month might now get 6,000. But it might also be feeding AI answers to tens of thousands of additional users who never visit the site at all. Those users are absorbing your framing, your language, your definitions, and often your organization’s name as the authority on the subject.

For an issue-focused organization, that kind of influence can matter more than raw traffic. If a major AI tool consistently describes your issue area using your language, cites your research, and names your organization as a credible source, you are shaping the public conversation in ways that are hard to match through traditional traffic alone.

The implication for strategy: do not abandon informational content. Double down on making it the best in your category. The goal shifts from “drive clicks to this page” to “be the source AI systems reach for when answering questions in our space.”

This also suggests new internal KPIs worth tracking. Brand mentions in AI answers. Citation frequency. Accuracy of how your organization is described. We cover how to track these below.

How Google AI Overviews differ from tools such as ChatGPT and Claude

These platforms all use generative AI, but they work in meaningfully different ways. Understanding the difference matters because optimizing for one does not automatically mean you are optimized for the other.

Google AI Overviews

AI Overviews are the summary boxes that appear above the traditional blue links in Google search results. They are an extension of Google’s core search product. A few characteristics:

• They are tightly connected to traditional search rankings. Ahref found that roughly 76% of URLs cited in AI Overviews also rank in the top 10 of the same Google search.
• The user is still on Google. They ran a search and got a summary with citations. The interaction model is search-first.
• Traditional SEO fundamentals (authority, backlinks, schema, on-page quality) still heavily influence whether your content shows up.

AI tools like ChatGPT, Claude, and Perplexity

These are conversational interfaces. A user opens a chat window, types or speaks a question, and gets a synthesized answer. They differ from AI Overviews in a few important ways:

• They often pull from a much wider and different pool of sources. Research from Brandlight suggests the overlap between Google’s top-ranking pages and the sources cited by ChatGPT has dropped below 20% on many queries. A page that does not rank on page one of Google can still get cited by an LLM.
• Each platform has its own preferences. Perplexity weighs recent content and heavily cites Reddit, YouTube, and established reference sites, per Exposure Ninja’s analysis. Claude tends to synthesize across sources and favors coherent, well-explained passages. ChatGPT often lifts bulleted and FAQ-style content directly. Gemini pulls heavily from Google’s existing index.
• Responses are non-deterministic. Ask Claude or ChatGPT the same question five times and you will see five slightly different answers with different sources. There is no “ranking” to check.
• Users are often further along in their decision-making and more likely to convert. A study from Superprompt found that AI search traffic converts at 14.2% compared to 2.8% for traditional Google traffic, though this will vary significantly by industry.

A useful way to think about it: AI Overviews are an evolution of search. ChatGPT, Claude, and Perplexity are a new channel entirely.

What AI referral traffic actually looks like

Digital Applied’s 2026 analysis found that ChatGPT sends roughly 190 times less referral traffic per query than Google. But because overall AI usage is growing so quickly, the absolute volume is rising fast. AI platforms generated over 1.1 billion referral visits in June 2025, a 357% jump from a year earlier.

For most of our clients, AI referral traffic is still a small slice of the total, typically 1-3% of all sessions. But it is growing month over month, and the users who do click through tend to be more engaged.

What AI systems look for when generating answers

Before getting to the practical optimization steps, it helps to understand how these systems work under the hood. Three concepts matter most:

• Passage-based retrieval. AI systems do not evaluate your page as a whole. They pull specific sections and evaluate each one on its own. A strong FAQ entry at the very bottom of a long page can get cited even if the rest of the page is weak.
• Multi-source synthesis. A single answer is typically compiled from several sources. Attribution varies by platform, and some of the content that shaped the answer may not be cited at all.
• Query fan-out. AI systems break broad questions into related subtopics and search for information on each. Content that covers a topic comprehensively performs better than content narrowly targeting a single keyword. We cover this in more detail below, since it has big implications for how you structure your site.

Some practitioners call the overall discipline Answer Engine Optimization (AEO). Others use Generative Engine Optimization (GEO) or LLM SEO. The terminology is still shaking out. The underlying work is largely the same.

How to track AI performance (and its limits

It is possible to track your performance in AI engines, though the tools are newer and the measurement is messier than traditional SEO. You will not get the equivalent of “rank #3 for this keyword.” But you can get a reasonable picture of how visible your organization is across AI platforms, and that picture is getting clearer every quarter.

There are three layers worth tracking.

1. Referral traffic

This is the easiest to track. Your existing analytics platform (Google Analytics 4, Plausible, Fathom, or whatever you are using) will show traffic from AI tools as referrals. Filter for sources including chatgpt.com, perplexity.ai, claude.ai, copilot.microsoft.com, and gemini.google.com.

Set up a dedicated segment or report for AI referral traffic so you can track growth over time. For most of our clients, the volume is still small but the trajectory is up.

2. AI Overview and AI citation tracking tools

This is where dedicated tools start to matter. We use Semrush on most of our client work. Its AI Overview tracking shows when your pages appear in AI Overviews for specific keywords, the estimated traffic impact, and how your presence is changing over time. Semrush also has a brand monitoring feature that tracks mentions across AI responses.

Other tools worth knowing about:

• SE Ranking offers similar AI Overview tracking and has been publishing good original research on AI citation patterns.
• Otterly.ai focuses specifically on AI search visibility across ChatGPT, Perplexity, Google AI Overviews, and others. Lighter weight than Semrush if you only need AI tracking.
• Profound and AthenaHQ are newer, AI-native tracking tools built specifically for generative engine visibility.
• Ahrefs has added AI Overview tracking to its Brand Radar feature and offers solid citation data if you already use it.

Most of these tools overlap in features. If you are already using Semrush or Ahrefs for SEO, starting with their AI modules makes sense. If you want something more focused, Otterly or Profound are good options.

3. Manual prompt testing

Underrated and often more useful than the tools above. Pick 10-20 prompts that matter to your organization and run them across ChatGPT, Claude, Gemini, and Perplexity on a regular cadence (monthly is a good baseline). For each prompt, note:

• Is your organization mentioned?
• Is your content cited as a source?
• How is your organization or issue area characterized?
• Is the characterization accurate?
• Which competitors or peer organizations show up instead of or alongside you?

This is a cheap, high-signal check that no tool fully replicates. If you are a nonprofit focused on affordable housing, periodically asking Claude “what are the most effective nonprofits working on affordable housing in the US” tells you more about your AI visibility than a dashboard will.

You can build a simple tracking spreadsheet with the prompts, the platforms, and the dates. Over time, you can see patterns emerge. Which prompts consistently include you. Which ones never do. Where you are being described accurately and where the AI is getting something wrong (which sometimes points to a messaging gap on your own site)

A note on the limits of these tools

Be realistic about what AI tracking tools can actually tell you. They are useful, but the data has real weaknesses and should be treated as directional rather than definitive.

A few things worth understanding:

• The data is sampled, not comprehensive. These tools do not see every AI response across every platform. They query a sample of prompts and platforms on a defined cadence, then extrapolate. Two different tools looking at the same organization can report meaningfully different numbers because they are sampling different prompts.
• Results are non-deterministic. Because AI systems generate different responses to the same prompt each time, a tool that checks your citation rate today will get a different answer tomorrow, even if nothing on your site has changed.
• Coverage across platforms is uneven. Some tools track ChatGPT and Google AI Overviews well but have limited coverage of Claude or Perplexity. Check what each tool actually monitors before you rely on the numbers.
• Attribution is incomplete. Even when an AI response is shaped by your content, it may not be cited. The tools can only track visible citations, which understates your real influence.

The practical takeaway: use these tools to spot trends over quarters, not to validate week-to-week changes. Combine the dashboard data with manual prompt testing to get a fuller picture. And do not expect the same precision you get from traditional SEO tools. The field is three years old. It will get better.

How to optimize for AI visibility

The good news is that optimizing for AI is mostly an extension of doing the basics well. There is no magic schema tag or secret trick. The sites that show up in AI answers are generally the ones that are clear, credible, current, and easy to parse.

Here is the approach we recommend to clients.

1. Make sure AI crawlers can actually access your site

This sounds obvious, but a surprising number of organizations accidentally block AI bots. Check your robots.txt, your CDN, and your firewall settings for blocks against GPTBot, ChatGPT-User, PerplexityBot, ClaudeBot, Google-Extended, and similar user agents. You should also avoid heavy paywalls, aggressive JavaScript rendering, or noindex tags on pages you want surfaced. Clean, server-rendered HTML improves retrieval.

Some organizations intentionally block AI crawlers for intellectual property reasons. That is a legitimate choice, but it has a cost. You cannot be cited by a system that cannot read your content. A Fuel Online industry report found that roughly a third of B2B SaaS companies are blocking AI crawlers, which effectively removes them from consideration in AI answers.

2. Write content that directly answers real questions

AI systems are trying to generate direct answers. They favor content that does the same. When we work with clients on content strategy, we push for:

• Clear, specific page titles that match how people actually phrase questions.
• An answer to the core question in the first paragraph, followed by supporting detail.
• Descriptive H2 and H3 subheadings. “Our Approach” tells an AI tool nothing. “How we evaluate grant applications” tells it exactly what the section is about.
• Short paragraphs, bullet points, and tables where they help. These formats are easier for AI to extract. Princeton’s GEO research found that content with statistics, citations, and quotations achieves 30-40% higher visibility in AI responses.

3. Write sections that can stand on their own

Because of passage-based retrieval, each section of a page needs to make sense without the rest of the page around it. That means avoiding constructions like “as we discussed above” or “see the previous section.” It means defining key terms where they appear, not just in a glossary elsewhere on the site. It means that a 150-word section answering a single specific question is often more valuable than a 2,000-word page that covers everything but forces the reader (and the AI) to do the synthesis work.

A good test: if an LLM needed a single 40-word passage from your page to answer a user’s question, would one exist? If not, write one.

4. Cover topics comprehensively, not just keywords

Because AI systems fan queries out into related subtopics, comprehensive coverage wins. A page that targets a single keyword but ignores the surrounding questions users actually ask is less likely to be cited than content that addresses the full topic.

The practical implication is to map out the real questions your audience has about a given topic and make sure your site answers each one clearly, whether that is on one thorough page or across a cluster of related pages. The goal is to be the organization that has covered the topic most credibly, not the one that has targeted the most keywords.

5. Invest in original data, research, and examples

AI systems weight original, specific, verifiable information heavily. A page that says “recurring donations are growing” will not get cited. A page that says “our recurring giving revenue grew 34% in FY25, representing 22% of total individual giving” very well might.

For a lot of organizations, this is actually a strength. You have benchmark reports, program data, survey results, and case studies nobody else has. SE Ranking’s analysis of 2.3 million pages identified content depth and original information as among the strongest predictors of AI citation, ahead of traditional SEO signals like backlinks. Publishing your data on your site in a clear, structured format is one of the single best things you can do for AI visibility.

6. Use structured data

Schema markup (Article, FAQ, HowTo, Organization, Event, Person) helps both search engines and AI systems understand your content. BrightEdge has reported that sites implementing structured data and FAQ blocks saw a 44% increase in AI search citations. Schema will not single-handedly get you cited, but it removes friction and is table stakes at this point. Our team implements schema as part of every site build.

Consistency matters too. Use the same organization name, descriptions, and entity references across your site so AI systems can confidently identify who you are.

7. Keep important content fresh

AI systems have a strong recency bias. BrightEdge found that pages updated within 60 days are 1.9 times more likely to appear in AI answers than older pages. This does not mean rewriting everything constantly. It means identifying your most important pages (the ones that would drive real value if cited) and reviewing them on a regular cadence. Update statistics. Replace dated examples. Add new developments. A blog post from 2022 about “the state of digital fundraising” is unlikely to get picked up in 2026.

8. Build authority the same way you always have

AI systems use many of the same trust signals Google does, and much of the E-E-A-T framework (Experience, Expertise, Authoritativeness, Trustworthiness) applies directly. Backlinks from credible sources, mentions in reputable publications, and a presence where your community discusses your issues (including Reddit, YouTube, and industry forums) all make it more likely that AI systems will trust and cite your content. SE Ranking found that pages mentioned on Reddit earn an average of 5.5 AI citations, compared to much lower rates for pages without community validation.

For associations, this often means making sure your research and reports are covered by the trade press. For nonprofits, it often means ensuring your program impact data shows up in news coverage, academic citations, and partner sites.

What to stop worrying about

A few things we see clients worry about that we think are distractions:

• Trying to “trick” AI systems. Prompt injection tricks, hidden text, and keyword stuffing either do not work or actively hurt you. The systems are too sophisticated and the rules are evolving too quickly.
• Obsessing over every platform. ChatGPT is the dominant AI tool by a wide margin. Semrush data puts its share of the AI chatbot market at roughly 80%. Optimize for it and Google AI Overviews first. The rest will largely follow.
• Panicking about SEO being dead. Traditional search is not going away, and the fundamentals of good SEO (clear content, technical health, authority) are the same fundamentals that drive AI visibility. If you have been doing SEO well, you are in a better position than you probably think.
• Treating traffic loss as the whole story. As noted earlier, the informational pages losing the most traffic are often the same pages powering AI responses. That influence is real, even when it does not show up as a site visit.

The bottom line

AI is changing how people find information online, and the effects on website traffic are real. But for organizations that have been investing in clear, well-structured, authoritative content, the transition is more of an evolution than a disruption.

The work is largely the same work. Answer real questions clearly. Publish original information your audience cannot get elsewhere. Write sections that can stand on their own. Keep your most important pages current. Make sure the technology on your site is sound and that AI crawlers can read your content. Track what you can, and use manual prompt testing to fill in the gaps.

And keep perspective on what success looks like. For nonprofits and associations especially, shaping the AI conversation about your issue area may ultimately matter more than the clicks you lose along the way.

If you want to talk through what this looks like for your site, reach out. We are happy to take a look at where you stand and what the highest-leverage improvements would be.