Search has never stood still. Every few years, a new layer gets added to how people find and evaluate information. Generative AI systems like ChatGPT, Copilot Search, and Perplexity haven’t replaced Google or Bing. They’ve added a new surface where discovery happens earlier, and where your visibility may never show up in analytics.
Call it Generative Engine Optimization, call it AI visibility work, or just call it the next evolution of SEO. Whatever the label, the work is already happening. SEO practitioners are already tracking citations, analyzing which content gets pulled into AI responses, and adapting strategies as these platforms evolve weekly.
This work doesn’t replace SEO, rather it builds on top of it. Think of it as the “answer layer” above the traditional search layer. You still need structured content, clean markup, and good backlinks, among the other usual aspects of SEO. That’s the foundation assistants learn from. The difference is that assistants now re-present that information to users directly inside conversations, sidebars, and app interfaces.
If your work stops at traditional rankings, you’ll miss the visibility forming in this new layer. Tracking when and how assistants mention, cite, and act on your content is how you start measuring that visibility.
Your brand can appear in multiple generative answers without you knowing. These citations don’t show up in any analytics tool until someone actually clicks.
Image Credi: Duane ForresterThe Missing Visibility Layer
Perplexity explains that every answer it gives includes numbered citations linking to the original sources. OpenAI’s ChatGPT Search rollout confirms that answers now include links to relevant sites and supporting sources. Microsoft’s Copilot Search does the same, pulling from multiple sources and citing them inside a summarized response. And Google’s own documentation for AI overviews makes it clear that eligible content can be surfaced inside generative results.
Each of these systems now has its own idea of what a “citation” looks like. None of them report it back to you in analytics.
That’s the gap. Your brand can appear in multiple generative answers without you knowing. These are the modern zero-click impressions that don’t register in Search Console. If we want to understand brand visibility today, we need to measure mentions, impressions, and actions inside these systems.
But there’s yet another layer of complexity here: content licensing deals. OpenAI has struck partnerships with publishers including the Associated Press, Axel Springer, and others, which may influence citation preferences in ways we can’t directly observe. Understanding the competitive landscape, not just what you’re doing, but who else is being cited and why, becomes essential strategic intelligence in this environment.
Why Mentions And Actions Matter
In traditional SEO, impressions and clicks tell you how often you appeared and how often someone acted. Inside assistants, we get a similar dynamic, but without official reporting.
- Mentions are when your domain, name, or brand is referenced in a generative answer.
- Impressions are when that mention appears in front of a user, even if they don’t click.
- Actions are when someone clicks, expands, or copies the reference to your content.
These are not replacements for your SEO metrics. They’re early indicators that your content is trusted enough to power assistant answers.
If you read last week’s piece, where I discussed how 2026 is going to be an inflection year for SEOs, you’ll remember the adoption curve. During 2026, assistants are projected to reach around 1 billion daily active users, embedding themselves into phones, browsers, and productivity tools. But that doesn’t mean they’re replacing search. It means discovery is happening before the click. Measuring assistant mentions is about seeing those first interactions before the analytics data ever arrives.
Traditional SEO Still Matters Most
Let’s be clear. Traditional search is still the main driver of traffic. Google handles over 3.5 billion searches per day. In May 2025, Perplexity processed 780 million queries in a full month. That’s roughly what Google handles in about five hours.
The data is unambiguous. AI assistants are a small, fast-growing complement, not a replacement (yet).
But if your content already shows up in Google, it’s also being indexed and processed by the systems that train and quote inside these assistants. That means your optimization work already supports both surfaces. You’re not starting over. You’re expanding what you measure.
From Rankings To Retrieval
Search engines rank pages. Assistants retrieve chunks.
Ranking is an output-aligned process. The system already knows what it’s trying to show and chooses the best available page to match that intent. Retrieval, on the other hand, is pre-answer-aligned. The system is still assembling the information that will become the answer and that difference can change everything.
When you optimize for ranking, you’re trying to win a slot among visible competitors. When you optimize for retrieval, you’re trying to be included in the model’s working set before the answer even exists. You’re not fighting for position as much as you’re fighting for participation.
That’s why clarity, attribution, and structure matter so much more in this environment. Assistants pull only what they can quote cleanly, verify confidently, and synthesize quickly.
When an assistant cites your site, it’s doing so because your content met three conditions:
- It answered the question directly, without filler.
- It was machine-readable and easy to quote or summarize.
- It carried provenance signals the model trusted: clear authorship, timestamps, and linked references.
Those aren’t new ideas. They’re the same best practices SEOs have worked with for years, just tested earlier in the decision chain. You used to optimize for the visible result. Now you’re optimizing for the material that builds the result.
One critical reality to understand: citation behavior is highly volatile. Content cited today for a specific query may not appear tomorrow for that same query. Assistant responses can shift based on model updates, competing content entering the index, or weighting adjustments happening behind the scenes. This instability means you’re tracking trends and patterns, not guarantees (not that ranking was guaranteed, but they are typically more stable). Set expectations accordingly.
Understanding Query Intent And Citation Potential
Not all content has equal citation potential, and understanding this helps you allocate resources wisely. Assistants excel at informational queries (”how does X work?” or “what are the benefits of Y?”). They’re less relevant for transactional queries like “buy shoes online” or navigational queries like “Facebook login.”
If your content serves primarily transactional or branded navigational intent, assistant visibility may matter less than traditional search rankings. Focus your measurement efforts where assistant behavior actually impacts your audience and where you can realistically influence outcomes.
How To Capture Mentions
The simplest way to start is manual testing.
Run prompts that align with your brand or product, such as:
- “What is the best guide on [topic]?”
- “Who explains [concept] most clearly?”
- “Which companies provide tools for [task]?”
Use the same query across ChatGPT Search, Perplexity, and Copilot Search. Document when your brand or URL appears in their citations or answers.
Log the results. Record the assistant used, the prompt, the date, and the citation link if available. Take screenshots. You’re not building a scientific study here; you’re building a visibility baseline.
Once you’ve got a handful of examples, start running the same queries weekly or monthly to track change over time.
You can even automate part of this. Some platforms now offer API access for programmatic querying, though costs and rate limits apply. Tools like n8n or Zapier can capture assistant outputs and push them to a Google Sheet. Each row becomes a record of when and where you were cited. (To be fair, it’s more complicated than 2 short sentences make it sound, but it’s doable by most folks, if they’re willing to learn some new things.)
This is how you can create your first “ai-citation baseline“ report if you’re willing to just stay manual in your approach.
But don’t stop at tracking yourself. Competitive citation analysis is equally important. Who else appears for your key queries? What content formats do they use? What structural patterns do their cited pages share? Are they using specific schema markup or content organization that assistants favor? This intelligence reveals what assistants currently value and where gaps exist in the coverage landscape.
How To Estimate Impressions
We don’t have official impression data yet, but we can infer visibility.
- Look at the types of queries where you appear in assistants. Are they broad, informational, or niche?
- Use Google Trends to gauge search interest for those same queries. The higher the volume, the more likely users are seeing AI answers for them.
- Track assistant responses for consistency. If you appear across multiple assistants for similar prompts, you can reasonably assume high impression potential.
Impressions here don’t mean analytics views. They mean assistant-level exposure: your content seen in an answer window, even if the user never visits your site.
How To Track Actions
Actions are the most difficult layer to observe, but not because assistant ecosystems hide all referrer data. The tracking reality is more nuanced than that.
Most AI assistants (Perplexity, Copilot, Gemini, and paid ChatGPT users) do send referrer data that appears in Google Analytics 4 as perplexity.ai / referral or chatgpt.com / referral. You can see these sources in your standard GA4 Traffic Acquisition reports. (useful article)
The real challenges are:
Free-tier users don’t send referrers. Free ChatGPT traffic arrives as “Direct” in your analytics, making it impossible to distinguish from bookmark visits, typed URLs, or other referrer-less traffic sources. (useful article)
No query visibility. Even when you see the referrer source, you don’t know what question the user asked the AI that led them to your site. Traditional search gives you some query data through Search Console. AI assistants don’t provide this.
Volume is still small but growing. AI referral traffic typically represents 0.5% to 3% of total website traffic as of 2025, making patterns harder to spot in the noise of your overall analytics. (useful article)
Here’s how to improve tracking and build a clearer picture of AI-driven actions:
- Set up dedicated AI traffic tracking in GA4. Create a custom exploration or channel group using regex filters to isolate all AI referral sources in one view. Use a pattern like the excellent example in this Orbit Media article to capture traffic from major platforms ( ^https:\/\/(www\.meta\.ai|www\.perplexity\.ai|chat\.openai\.com|claude\.ai|gemini\.google\.com|chatgpt\.com|copilot\.microsoft\.com)(\/.*)?$ ). This separates AI referrals from generic referral traffic and makes trends visible.
- Add identifiable UTM parameters when you control link placement. In content you share to AI platforms, in citations you can influence, or in public-facing URLs. Even platforms that send referrer data can benefit from UTM tagging for additional attribution clarity. (useful article)
- Monitor “Direct” traffic patterns. Unexplained spikes in direct traffic, especially to specific landing pages that assistants commonly cite, may indicate free-tier AI users clicking through without referrer data. (useful article)
- Track which landing pages receive AI traffic. In your AI traffic exploration, add “Landing page + query string” as a dimension to see which specific pages assistants are citing. This reveals what content AI systems find valuable enough to reference.
- Watch for copy-paste patterns in social media, forums, or support tickets that match your content language exactly. That’s a proxy for text copied from an assistant summary and shared elsewhere.
Each of these tactics helps you build a more complete picture of AI-driven actions, even without perfect attribution. The key is recognizing that some AI traffic is visible (paid tiers, most platforms), some is hidden (free ChatGPT), and your job is to capture as much signal as possible from both.
Proxies For Machine-Validated Authority
Machine-Validated Authority (MVA) isn’t visible to us as it’s an internal trust signal used by AI systems to decide which sources to quote. What we can measure are the breadcrumbs that correlate with it:
- Frequency of citation
- Presence across multiple assistants
- Stability of the citation source (consistent URLs, canonical versions, structured markup)
When you see repeat citations or multi-assistant consistency, you’re seeing a proxy for MVA. That consistency is what tells you the systems are beginning to recognize your content as reliable.
Early Benchmarks And What They Mean
Perplexity reports almost 10 billion queries a year across its user base. That’s meaningful visibility potential even if it’s small compared to search.
Microsoft’s Copilot Search is embedded in Windows, Edge, and Microsoft 365. That means millions of daily users see summarized, cited answers without leaving their workflow.
Google’s rollout of AI Overviews adds yet another surface where your content can appear, even when no one clicks through. Their own documentation describes how structured data helps make content eligible for inclusion.
Each of these reinforces a simple truth: SEO still matters, but it now extends beyond your own site.
Building A Tracking Stack
Start small. A basic spreadsheet is enough.
Columns:
- Date.
- Assistant (ChatGPT Search, Perplexity, Copilot).
- Prompt used.
- Citation found (yes/no).
- URL cited.
- Competitor citations observed.
- Notes on phrasing or ranking position.
Add screenshots and links to the full answers for evidence. Over time, you’ll start to see which content themes or formats surface most often.
If you want to automate, set up a workflow in n8n that runs a controlled set of prompts weekly and logs outputs to your sheet. Even partial automation will save time and let you focus on interpretation, not collection. Use this sheet and its data to augment what you can track in sources like GA4.
Cost-Benefit Reality Check
Before investing heavily in assistant monitoring, consider resource allocation carefully. If assistants represent less than 1% of your traffic and you’re a small team, extensive tracking may be premature optimization. Focus on high-value queries where assistant visibility could materially impact brand perception or capture early-stage research traffic that traditional search might miss.
Manual quarterly audits may suffice until the channel grows to meaningful scale. This is about building baseline understanding now so you’re prepared when adoption accelerates, not about obsessive daily tracking of negligible traffic sources.
Reporting Internally
Executives understand and prefer dashboards, not debates about visibility layers, so show them real-world examples. Put screenshots of your brand cited inside ChatGPT or Copilot next to your Search Console data. Explain that this is not a new algorithm update but a new front end for existing content. It’s up to you to help them understand this critical difference.
Frame it as additive reach. You’re showing leadership that the company’s expertise is now visible in new interfaces before clicks happen. That reframing keeps support for SEO strong and positions you as the one tracking the next wave.
Legal And Ethical Context
It’s worth noting that citation practices exist within a shifting legal landscape. Publishers and content creators have raised concerns about copyright and fair use as AI systems train on and reproduce web content. Some platforms have responded with licensing agreements, while legal challenges continue to work through courts.
This environment may influence how aggressively platforms cite sources, which sources they prioritize, and how they balance attribution with user experience. The frameworks we build today should remain flexible as these dynamics evolve and as the industry establishes clearer norms around content usage and attribution.
What The Data Really Tells Us
AI assistant visibility is not yet a major traffic source. It’s a small but growing signal of trust.
By measuring mentions and citations now, you build an early-warning system. You’ll see when your content starts appearing in assistants long before any of your analytics tools do. This means that when 2026 arrives and assistants become a daily habit, you won’t be reacting to the curve. You’ll already have data on how your brand performs inside these new systems.
If you extend the concept here of “data” to a more meta level, you could say it’s already telling us that the growth is starting, it’s explosive, and it’s about to have an impact in consumer’s behaviors. So now is the moment to take that knowledge and focus it on the more day-to-day work you do and start to plan for how those changes impact that daily work.
Closing Thought
Traditional SEO remains your base layer. Generative visibility sits above it. Machine-Validated Authority lives inside the systems. Watching mentions, impressions, and actions is how we start making what’s in the shadows measurable.
We used to measure rankings because that’s what we could see. Today, we can measure retrieval for the same reason. This is just the next evolution of evidence-based SEO. Ultimately, you can’t fix what you can’t see. We cannot see how trust is assigned inside the system, but we can see the outputs of each system.
The assistants aren’t replacing search (yet). They’re simply showing you how visibility behaves when the click disappears. If you can measure where you appear in those layers now, you’ll know when the slope starts to change and you’ll already be ahead of it.
More Resources:
- How LLMs Interpret Content: How To Structure Information For AI Search
- Introduction To Vector Databases And How To Use AI For SEO
- The State Of AI In Marketing
Featured Image: Anton Vierietin/Shutterstock
This post was originally published on Duane Forrester Decodes.
