THE AI TRAFFIC SQUEEZE: HOW PERFORMANCE MARKETERS CAN TURN SCRAPED DATA INTO A COMPETITIVE EDGE

AI isn’t just “changing search.” It’s quietly constricting the very oxygen performance marketers live on: impressions and clicks.

You’re feeling this as channel-level symptoms—Google Ads CPCs inching up, branded search campaigns getting fewer impressions, organic traffic eroding even when rankings look “fine.” Underneath those numbers is a structural shift: AI systems are intercepting demand, answering more queries themselves, and passing far fewer visitors through to your site.

On the paid side, Google has literally shrunk the canvas. AI Overviews and AI Mode now occupy the prime real estate where multiple ad units used to live, so there are simply fewer impression opportunities per query. In one advertiser’s account, Search Engine Journal’s analysis showed an 11% year-over-year decline in impressions, even as demand for the product didn’t change. That’s not seasonality; it’s structural compression.

Yes, Google is experimenting with running ads inside AI Mode and rolling out new formats like Direct Offers, but even those formats sit in a smaller, more constrained environment. Instead of a scrollable page of ad units and organic results, you’re fighting for presence inside a compact AI answer where the bar for inclusion is much higher—and the user often doesn’t see anything else.

At the same time, the “audience” you’re marketing to is evolving from human searchers to a mix of people and agents. Microsoft describes this as three overlapping eras of the web: “help me find it,” “help me choose,” and “do it for me.” That last mode—“do it for me”—is growing fastest, fueled by automated traffic that Microsoft says is rising roughly eight times faster than human visits. When the user’s first step is “Copilot, book this” or “Gemini, pick the best option and check out,” your campaign isn’t selling to a person scrolling a SERP; it’s competing to be the option an agent silently selects.

Agents don’t scroll, and they don’t respond to a clever preheader or hero image. They evaluate, shortlist, and act. Retail media strategist Roger Dunn calls the result the “shortlist economy”: ask ChatGPT, Gemini, or Copilot for a recommendation and you typically get three to five options back. That shortlist becomes the entire consideration set. According to a December 2025 survey cited in Search Engine Journal’s piece on agentic commerce, 43% of U.S. shoppers had already discovered a new brand via AI, and 47% said they notice AI-mentioned brands often or very often. If you’re not in that three-to-five result set, your brand is effectively invisible in that buying moment.

Organic performance is under the same pressure. As AI Overviews and other generative panels answer more questions directly on the results page, the clicks you used to bank on are disappearing. Data from Ahrefs’ coverage of AI search trends shows a surge in interest around “zero-click search strategy” and “why is my website traffic dropping,” reflecting how marketers have shifted from vague concern to active diagnosis of traffic loss. But fixing rankings alone no longer fixes revenue. As MarTech’s breakdown of Google’s updated content guidance points out, a page can improve in quality—and even in Google’s internal assessment—without regaining its former traffic, because AI Overviews, forums, and video now occupy more of the SERP. A perfectly rehabilitated article cannot reclaim a click that Google (or an AI layer) no longer sends.

Meanwhile, AI systems are voraciously scraping the open web to power these experiences. An analysis from the WordStream team, who examined millions of AI bot hits, found that bots were relentlessly crawling their content while sending back a “measly” amount of traffic. Their blog had already lost 85% of organic visits over four years, and those scrapes often reproduced first-party data and insights directly inside AI answers. That’s the new bargain: more of your data is being ingested, while less of the resulting demand makes its way back to you.

Put these pieces together and the “impression squeeze” isn’t just about having fewer ad slots or lower organic CTR. It’s about a world where:

• A growing share of queries are satisfied by AI on the results page.
• Agentic systems compress the consideration set into a tiny shortlist.
• Bots aggressively harvest your content while returning marginal traffic.

In this environment, the unit economics of performance marketing change. Every click you do win is more expensive, more contested, and more valuable—and the untracked shadow audience of AI agents is shaping demand long before a human ever lands on your site.

“Optimizing for AI Overviews” sounds like the natural next frontier for SEO. In reality, it’s a trap for performance marketers, for the same reason “chasing blue links” turned into a treadmill: you end up optimizing for the interface instead of the buyer.

The first problem is structural. AI Overviews and agentic experiences are not neutral new placements; they’re a traffic dam. As one analysis of Google’s latest content guidance notes, even if you improve weak content and recover rankings, you’re still competing in a results page where AI modules, forums, and video “take up more space,” which means “a page can get better without getting its clicks back” because Google now answers more of the query itself before anyone visits your site, as MarTech explains. Scrambling to “win” a citation in an AI Overview doesn’t change the fact that the Overview exists to reduce outbound clicks.

The second problem is that AI surfaces don’t behave like classic search listings. In traditional search, a query like “best yoga mats for home workouts” yields a buffet of ranked links; users scan options and decide where to click. In AI assistants, the same question becomes a compressed recommendation, a handful of brands in a single synthesized answer. The brand either appears in that answer or it doesn’t, and the user often never sees an open marketplace of results, as the HubSpot marketing team describes when contrasting search with answer engines. Trying to “optimize” for this the way you would for a rich snippet misreads the interface; you’re no longer nudging a click—you’re fighting to exist on a shortlist.

That shortlist dynamic is exactly what’s playing out across paid media. As performance advertisers are discovering, AI Overviews and “AI Mode” have already eaten into the real estate where ads used to appear, driving an 11% year-over-year decline in impressions in one Search Engine Journal study. In the words of retail strategist Roger Dunn, we’ve entered a “shortlist economy,” where agents return three to five options and that set becomes the whole consideration universe. When you make “optimize for AI Overviews” your goal, you’re effectively optimizing for a brutally tight gatekeeper that was designed to reduce the number of decisions a user has to make.

The third, subtler trap is that AI-Overview optimization pushes teams toward the exact behavior Google and other platforms are now actively penalizing: scaled, low-value content. Because AI feels like a content vending machine, the temptation is to flood the web with pages tailored to hypothetical prompts and query variations. But Google’s updated abuse guidance explicitly flags using generative AI to create large numbers of pages “without adding value,” and stresses that there must be a human editor with enough context and authority to improve or kill weak work before it goes live, as MarTech’s analysis makes clear. The fastest way to disappear from AI Overviews entirely is to crank out AI-generated fluff aimed at manipulating them.

There’s also a strategic cost: every cycle your team spends reverse-engineering Gemini’s wording is a cycle not spent building the assets AI systems are actually hungry for—credible signals of expertise, distinctive perspectives, and proof. When Lily Ray’s research on AI-assisted SEO emphasized that winning content still needs “unique insights, firsthand expertise, and original perspectives” that can’t be pasted from a model output, she was describing the kind of material that both traditional search and AI systems lift into their answers, as summarized on Neil Patel’s blog. That kind of content is slow to produce and hard to clone—which is exactly why it’s defensible.

Finally, “AI Overview optimization” is a measurement trap. It pulls you back into a traffic-first mindset just as the traffic-based funnel is breaking. The emerging AI discovery layer acts as a smart filter between users and your site, resolving many informational and comparative queries without a visit at all, which means raw organic sessions tell you less and less about whether your marketing is working, as MarTech argues. Treating AI Overviews as the new ranking report encourages you to chase visibility in a layer that’s explicitly designed to hide you.

For performance marketers, the opportunity is not to “rank” in AI Overviews. It’s to understand how and where AI systems already compress your category into shortlists—and then build the brand, data, and experience signals those systems can’t ignore. The next sections will focus on how to do that without falling back into the old traffic-chasing reflex.

Start by reframing what’s happening: AI isn’t just “stealing” your traffic—it’s also running a public, always-on focus group on your entire category. Every AI Overview, every ChatGPT recommendation, every agentic “do this for me” flow is a stitched-together composite of the content, offers, and positioning that are winning the most confidence from the model right now.

If you treat AI as a competitor you can spy on instead of a black box to appease, you can mine those composites for patterns—and then use them to sharpen your own performance strategy.

The first layer is simple visibility: who is showing up, where, and for what kind of prompts. AI search analytics tools now let you track which brands get surfaced, which pages they’re pulling from, and what queries trigger those mentions. As one breakdown of “answer engine optimization” tools explains, marketers are already using these platforms to see which competitors appear instead of them for high-intent prompts and to identify content gaps that aren’t covered by their current pages or ads, especially in research-heavy B2B journeys where AI-assisted buyers are now the norm.

That’s useful—but the performance edge comes from going a step deeper: reading these AI surfaces like a new kind of ad library.

When you look at a set of AI Overviews and ChatGPT answers for the same commercial topic over time, you’ll notice recurring ingredients:

• Product attributes the model consistently emphasizes (e.g., “lifetime warranty,” “SOC 2 compliance,” “30-day onboarding”).
• Offer structures it keeps recommending (“free migration,” “usage-based pricing,” “no annual contract”).
• Brand types it trusts in each niche (pure SaaS, agencies, marketplaces, aggregators, publishers).

The WordStream team, while testing the volatility of AI Overview citations, accidentally uncovered exactly this sort of signal. Their repeated Thursday queries produced rotating citations, but they also revealed strong clues about the kinds of content Google’s AI currently “likes” for marketing topics—patterns they only spotted because they logged the same 14 searches, week after week, and compared which sources and formats kept resurfacing in those.

You can run a similar cadence, but with a performance lens:

1. Define a test panel of prompts. Mirror your funnel: “best [category] tools,” “[category] pricing,” “[competitor] alternatives,” “[category] for [segment].”
2. Capture AI answers on a schedule. Screenshots or exports every week or month from Google, ChatGPT, and any vertical AI engines in your space.
3. Tag the components. For every cited competitor, log: positioning claims, proof points, CTAs, pricing language, and any recurring trust markers.
4. Roll up the patterns. Which value props and offer mechanics are overrepresented among the brands AI keeps resurfacing?

This is where scraped data turns into competitive advantage. AI systems are compressing the entire category’s messaging into a handful of phrases and proof points. If every AI answer in your niche emphasizes “no long-term contracts” and “transparent pricing,” that’s not just content bias—that’s a live readout of what the market narrative has converged around.

Overlay that with what you know from search behavior and bot logs. When you see a spike in AI agent hits on certain pages in tools like Ahrefs Bot Analytics, it’s a tell that those topics are being harvested more aggressively for AI answers. Pairing that with rising searches for things like “zero-click search strategy” and traffic-loss audits, which recent AI search trend analysis has documented, gives you a shortlist of thematic battlegrounds where AI is absorbing demand fastest.

Now you can make deliberate moves:

• Refine ad messaging using AI-era proof points. If AI answers for “IT asset management software” repeatedly highlight “agentless deployment” and “real-time visibility,” bake those phrases, or your defensible alternatives, into search and social ad copy. You’re co-opting the language buyers will already have seen in AI research.

• Engineer “must-cite” assets. Look at which competitor pages AI keeps pulling from. Are they in-depth comparisons? Pricing explainers? Implementation guides? Then build the stronger, more opinionated version anchored in your POV. As Lily Ray has argued about AI-supported content, pages that combine research with firsthand expertise and original perspective are harder for competitors (or models) to copy-paste their way around; they also tend to become the canonical sources AI leans on.
• Spot weak points in competitor funnels. If AI persistently recommends a rival for “best for enterprises” but never surfaces them for “pricing transparency” or “fast implementation,” that’s a segmentation map for your campaigns. Push performance spend into the gaps where AI’s composite picture of the category is underdeveloped and you can own the click even when you lose the mention.

• Anticipate offer inflation. When you start seeing more aggressive offers (“90-day free trial,” “unlimited seats”) showing up in AI answers, assume those promotions will soon be weaponized across paid channels. Decide in advance whether you’ll match, counter-position (“no gimmicky trial periods”), or outflank with a different axis of value.

The point isn’t to “optimize for the model.” It’s to use the model as a highly opinionated aggregator of what everyone else in your market is saying, shipping, and testing. AI might be compressing your traffic—but it’s also exposing your competitors’ moves in a form you can systematically mine, codify, and feed back into higher-converting creative, offers, and landing pages.

In a shortlist economy, the funnel doesn’t start at your ad or your homepage. It starts inside an AI-generated answer where a buyer sees three to five options — and either you’re on that list or you’re invisible. That’s a completely different design problem than the classic “awareness → consideration → conversion” journey most performance marketers are used to.

Two things are happening at once.

First, AI is compressing the journey. Agentic experiences are collapsing “help me research” and “help me decide” into a single interaction, which is why researchers have found AI-referred visitors convert at more than four times the rate of traditional organic search visitors, as one HubSpot analysis of AI search behavior points out. Second, Google and other platforms are packing more of that journey into the results page itself. AI Overviews, Direct Offers, product carousels, and community snippets mean far more evaluation now happens before a user ever clicks.

For marketers, that means your funnel has to assume two parallel realities:

• Journeys that begin and mostly unfold inside AI answers.
• Journeys that still begin in search, but where the SERP is now an opinionated, brand-heavy comparison layer you can’t skip.

You can’t fix this by “ranking higher” alone, because ranking doesn’t guarantee attention. A detailed look at how Google is reshaping SERPs notes that even “perfectly rehabilitated” content may not regain its former traffic when AI Overviews and rich results crowd the page; better pages don’t reclaim clicks Google no longer sends, as MarTech’s reporting on content guidance updates makes clear. In practice, that means you need funnels that assume fewer first-click opportunities and do more with every impression you get.

Start by redefining the top of the funnel. If AI is the new discovery layer, the first “touch” with your brand is often a mention inside a conversational answer that sends no traffic at all. Measurement teams are already responding by tracking brand demand — direct visits, branded search, and social chatter — as a proxy for AI-driven discovery, rather than fixating on raw sessions, a shift documented in recent MarTech guidance on measuring when AI owns discovery. Your funnel model should mirror that: treat branded search spikes and “who are these guys?” visits as the post-AI top of funnel, not as random direct traffic.

Next, design intent-specific “shortlist landers” instead of generic entry points. If a buyer arrives after asking “best X for Y” in an AI chat, they’ve effectively skipped half the consideration phase. Your landing experiences should acknowledge that:

• Lead with category context and differentiation, not definitions. AI already educated the buyer.
• Show comparison-friendly information above the fold: pricing logic, use cases, social proof, implementation details.
• Make switching or trial frictionlessly easy with clear “start now” or “test in my stack” paths.

Because AI surfaces are brand-heavy, your SERP presence must also work as a mini-landing page. In a world where impressions are shrinking — with one Search Engine Journal analysis of agentic commerce documenting double-digit declines in ad impressions as AI modes take over screen real estate — every pixel that still shows your brand has to communicate something real. That means:

• Titles and meta descriptions that carry your positioning, not just keywords.
• Schema, images, and reviews that visually anchor your brand as “the serious option” in a crowded panel.
• Offers that match the agent’s job to be done (configure, compare, book) rather than generic “learn more” CTAs.

Finally, shorten the middle of your funnel on purpose. If more evaluation is happening upstream, you don’t need a labyrinth of nurture touches; you need clear, high-intent paths once someone lands. Collapse overlapping content, route visitors quickly into self-serve demos or calculators, and reserve depth for the buyers who raise their hands. AI has already pre-qualified many of them — your job is to remove the remaining friction, not recreate the research they’ve already done elsewhere.

AI isn’t just squeezing your traffic; it’s silently enforcing a new standard of what “safe,” compliant, and credible messaging looks like in your category. Every bot scrape, every AI Overview, every ChatGPT recommendation is a feedback loop on which narratives the models trust enough to repeat. Ad intelligence is how you decode that loop.

Most teams still look at scraped data through a narrow lens: “Are our pages being used without sending us traffic?” That’s a valid concern, especially when you see crawlers consuming huge volumes of content and sending back a trickle of visits, as the team at WordStream found in their analysis of millions of AI bot scrapes. But performance marketers can flip this dynamic. Instead of just asking, “Is this worth it?”, you can ask, “What is the model learning about my category’s compliant, high‑trust messaging — and how do I weaponize that insight?”

Start with the actual answers AI is giving. Tools that surface where your brand appears in AI answers, which prompts trigger those mentions, and which sources are being cited — like the “track AI mentions, analyze citations, monitor prompts” workflows outlined in HubSpot’s overview of AI search analytics tools — give you a live map of how models describe your space. Pull that into your ad intelligence stack and classify what you see:

• Which competitors consistently show up in AI Overviews and conversational answers for high‑intent queries?
• What exact phrasing appears right before or after compliance‑sensitive topics (claims, guarantees, pricing, regulated benefits)?
• Which disclaimers, qualifiers, or “safety language” are the models preserving word‑for‑word?

Patterns here are not random. AI systems are heavily regularized toward content that looks factually defensible, brand‑consistent, and aligned with established expertise. When a competitor keeps appearing in “best for X” recommendations, their messaging is likely hitting a sweet spot of specificity without tripping risk filters. Instead of copying their tagline, you reverse‑engineer the attributes: how they frame proof, what kinds of numbers they use, whether they reference third‑party validation, how they word limitations.

This matters because AI‑referred visitors already convert at dramatically higher rates — HubSpot’s analysis puts it at 4.4x compared with traditional organic search. That suggests models are quietly curating not just relevance, but trust and fit. Your job is to understand what “trust and fit” looks like in the language they actually output and then reflect those same patterns, in your own voice, across ads, landing pages, and scripts.

Scrape data from your own properties helps you see what is being “canonized” by the bots today. WordStream’s finding that AI agents gravitate toward pages rich in first‑party data, research, and applied insights — then lift those elements almost verbatim in AI results — confirms what many SEOs have seen anecdotally: models lean hardest on content that feels like expert summary, not fluffy inspiration or generic how‑to material. When your analytics show that certain studies, frameworks, or process breakdowns are overrepresented in AI answers, treat those assets as templates for both:

• AI‑safe messaging: factual, nuanced, grounded in observable data.
• Compliance‑resilient creative: claims traceable to a clear methodology, with limitations and context baked in.

This is also where human review becomes non‑negotiable. As Neil Patel has noted in his discussion of human‑led SEO, brands that delegate too much to generic AI copy tend to drift away from their core identity, enjoy a short‑term traffic spike, and then get hammered when algorithms tighten quality requirements. The brands that keep winning have humans in the loop stress‑testing each message: Does this overpromise? Is this actually our POV? Would legal sign off? AI‑safe patterns are about more than “what the model repeats;” they’re about which messages can survive repeated reinterpretation by third‑party systems without becoming misleading.

Operationally, you can turn this into an ad‑intelligence practice:

1. Monitor AI citations and prompts weekly. Pull data from AI search analytics into the same dashboards you use for creative performance so planners and compliance see the same source of truth.
2. Tag recurring safety and compliance motifs. For each high‑intent topic, catalogue the standard caveats, eligibility statements, and proof structures that keep showing up in AI responses.
3. Build “AI‑safe” message templates. Convert those motifs into ready‑to‑deploy ad hooks and value props that already resemble what models are willing to echo — just customized to your brand’s voice and legal constraints.
4. Stress‑test creative against AI outputs. Before scaling a new messaging angle, ask the same question your audience would in a leading AI assistant. If your language is absent — and a competitor’s compliant framing is consistently present — you know which direction to adjust.

In the traffic squeeze, scraped data is the cost of doing business. Ad intelligence on that same data is the upside: a near‑real‑time x‑ray of which messages AI has decided are safe, compliant, and persuasive enough to repeat — and therefore which messages you should be scaling across every performance channel.