Google AI Overviews are eating your clicks. Here is what to do about it.

If your organic traffic dipped this year while your rankings held, you are not imagining it. The page is still ranking. The click is just no longer happening. Between the search box and your website, Google has inserted a new layer: an AI-written summary that answers the question on the spot, names a few sources, and quietly removes the reason most people used to click through. This is the AI Overviews era, and for a lot of businesses it shows up first as a line going the wrong way in Google Analytics.

The honest framing matters here, because most advice on this topic is either denial ("traffic is fine, just measure better") or panic ("SEO is dead"). Neither is useful. Some of your traffic is genuinely gone and is not coming back. Some of it was never going to convert anyway. And a meaningful slice of what remains is now more valuable than it was before. The job is to tell those three apart, defend the part worth defending, and reposition around the one thing that still compounds: being the source the AI cites.

This guide is long on purpose. It is the reference we wish existed when a founder forwards us a screenshot of a falling traffic chart and asks whether the sky is falling. We will define what an AI Overview actually is, separate the queries you are about to lose from the ones you will keep, lay out a step-by-step playbook for earning citations, give you a measurement framework that does not lie to you, catalog the mistakes that turn a manageable shift into a self-inflicted wound, and close with an FAQ you can hand to a skeptical executive. Read it once end to end, then keep it as a checklist.

What an AI Overview actually is

An AI Overview is the synthesized answer block Google places above the traditional results for a growing share of queries. Instead of returning ten blue links and letting you choose, Google reads several pages, composes a direct answer in its own words, and shows a short list of linked sources beside or beneath it. For the searcher it feels like the answer arrived first and the links became optional. That is precisely the problem for publishers and brands: the page that supplied the facts often does not get the visit.

This is the same shift we covered in our piece on GEO versus SEO, viewed from the traffic side of the ledger rather than the strategy side. Classic SEO asks "do I rank?" The new question is "am I inside the answer, and if a click happens at all, does it come to me?" Those are different questions with different answers, and your analytics is now measuring the gap between them.

Why Google built it, and why it is not going away

It helps to understand why Google did this, because the motive tells you it is not going away. Google was facing real competitive pressure from conversational assistants that answer in plain language instead of handing back a list to sift through. The Overview is Google answering in kind: keep the user inside Google, satisfy the question immediately, and defend the habit before a rival captures it. A feature built to protect the core franchise does not get rolled back because publishers complain about clicks. Plan as though the answer layer is permanent, because the incentive that created it is.

There is a second motive worth naming. Every query that resolves inside the Overview is a query that stays inside Google's surface, where Google controls the experience and, eventually, the monetization. The link economy was a bargain in which Google sent you traffic in exchange for crawling your content. The answer economy renegotiates that bargain in Google's favor. You can dislike the new terms, but you cannot opt out of them and remain visible. The realistic posture is to play the new game well rather than to litigate the old one.

How an Overview decides what to show

You do not get a peek inside the model, but the observable behavior is consistent enough to reason about. An Overview tends to appear on queries that are informational, that have a reasonably settled answer, and where Google is confident it can compose something useful without sending you elsewhere. It tends not to appear, or appears in a thinner form, on queries that are transactional, navigational, highly local, or genuinely contested. The sources it pulls skew toward pages that state a clean, self-contained answer early, that carry recognizable authority signals, and that the broader web already treats as credible on the topic. None of that is a secret formula. It is the same quality bar that has always rewarded good content, applied to a new surface and with a new payoff.

Zero-click is not new. The scale is.

Zero-click search predates AI by years. Weather, currency conversions, sports scores, "what time is it in Tokyo," public-figure birthdays: Google has answered these inline for a decade, and nobody mourned the lost clicks because those clicks were never commercially interesting. What AI Overviews changed is the type of query that now resolves without a visit. The summary box has moved up the funnel, into the explanatory and comparative questions where customers used to form opinions by reading your content.

So the useful mental model is not "clicks are disappearing." It is "the answer is being intercepted earlier, and on a wider range of questions than before." Whether that hurts you depends almost entirely on which questions your traffic came from.

Read click-through rate by intent, not as one number

The other thing that changed is click-through rate by query type, and it is uneven enough that an aggregate number will mislead you. On a query where the Overview fully answers the question, the click-through rate on the top organic result can fall sharply, because the searcher has what they came for before they reach the links. On a query where the Overview only frames the topic and names a few candidates, the picture is gentler, and being named in the box can even lift qualified clicks. Average those two together and you get a middling figure that describes neither. The practical move is to stop reading your site-wide click-through rate as one number and start reading it per query intent, because that is the only resolution at which the damage and the opportunity are actually visible.

A useful habit: export your Search Console queries, tag each one by intent (informational, comparative, transactional, branded), and compute click-through rate within each bucket month over month. The aggregate line will look like slow decline. The bucketed view will usually show one bucket collapsing, two holding, and one rising. That is the difference between knowing your traffic is "down" and knowing exactly which part of it left and why.

Which queries lose traffic, and which are safe

Not all of your keywords are exposed equally. It helps to sort your traffic into three buckets by what the searcher is actually trying to do. The table below is the version of this we sketch on a whiteboard in nearly every first conversation.

Highly exposed: expect real loss

• Definitional and "what is" queries. "What is generative engine optimization," "what does APR mean," "how does an air fryer work." These are exactly what the summary box was built to answer. If a large share of your sessions came from this kind of top-of-funnel explainer content, that traffic is the most likely to thin out.
• Simple how-to and quick-answer queries. "How to convert PDF to Word," "how many ounces in a cup," "how to reset a router." The answer fits in a paragraph, so the click is now optional. Recipe and "near me" snippets have lived with versions of this pressure for years; the difference is the range of topics it now touches.
• Single-fact lookups. Anything answerable in one line. If your blog won attention by being the cleanest source for a discrete fact, the AI can now lift that fact and rarely needs to send the visit.

Partly exposed: it depends on depth

• Comparison and "best X for Y" queries. The Overview may name two or three options, which can actually surface your brand if you are one of them, but it also satisfies the casual comparer who only wanted a shortlist. You lose the browser and keep the customer.
• "Versus" queries. "Shopify vs WooCommerce," "DTC vs marketplace." The summary handles the surface contrast. The searcher who needs the real tradeoffs, the kind we lay out in our breakdown of DTC versus platform costs, still clicks through for the depth a paragraph cannot hold.

Largely protected: mostly safe

• Transactional and bottom-of-funnel queries. "Buy," "pricing," "demo," "book a consultation," branded searches for your company name. People at the point of action want to reach a real page, complete a transaction, or talk to a human. A summary does not finish the job, so the click survives.
• High-stakes, high-consideration decisions. Anything involving meaningful money, legal or medical weight, or professional risk. Buyers do not commit to a procurement decision because a paragraph told them to; they want primary sources, methodology, and someone accountable for the claim.
• Genuinely original content. Proprietary data, first-party research, named case studies, opinionated analysis with a point of view. The AI can summarize the gist, but it has to attribute the source, and the people who care about the specifics go looking for the original.

The traffic most exposed to AI Overviews is the traffic that was least likely to buy. The traffic that survives is closer to your pipeline.

That last point is the one to sit with. The sessions you are losing skew toward casual, top-of-funnel, low-intent visits. Painful for a pageview chart, far less painful for a revenue chart. The work is to stop optimizing for the volume you cannot keep and start engineering for the citations and the high-intent clicks you can.

A worked example: the chart that screams vs. the chart that matters

A simplified, illustrative example makes the point concrete. Say a B2B site draws 50,000 organic sessions a month: 35,000 from broad informational and definitional content, 15,000 from product, pricing, and comparison pages. The informational traffic converts at a fraction of a percent; the bottom-of-funnel traffic carries almost all the pipeline. If AI Overviews shave 40% off the informational pool but barely touch the protected pages, total sessions drop by roughly 14,000, a number that looks alarming on a dashboard. Revenue barely moves, because the lost visits were the ones that almost never converted. The same business, judged by sessions, looks wounded; judged by pipeline, it is fine, and its remaining traffic is now denser and cheaper to serve. Those figures are illustrative, not a benchmark, but the shape is what we see repeatedly: the chart that screams and the chart that matters are telling different stories.

Here is the same scenario in numbers, so you can run the logic against your own analytics.

Sessions fell by about 30% in this illustration. Conversions fell by a small fraction, because the segment that collapsed was the one that barely converted in the first place. If your dashboard leads with sessions, you will sound an alarm that your revenue does not justify. If it leads with conversions per segment, you will see a business that is roughly intact and a content portfolio that needs rebalancing rather than rescuing. Most of the panic we are asked to talk founders down from is a measurement problem wearing a traffic problem's clothes.

How to become the source AI Overviews cite

When Google composes an Overview, it is choosing which sources to lean on and link. Getting chosen is not luck; it rewards specific, learnable properties in your content and your brand. We go deeper on the mechanics in how to get cited by AI, but here is the core of what earns the citation.

1. Answer the question cleanly in the first two sentences. AI engines extract self-contained, declarative passages. Lead with the direct answer, then explain. A point buried under 400 words of preamble rarely gets pulled into a summary. Write the way you would want to be quoted.
2. Be specific and attribute your numbers. "AI Overviews now appear on a large and growing share of informational queries (cite the study and year)" is far more quotable than "AI is changing search." A named figure with a named source reads as trustworthy and is easy for an engine to lift with attribution.
3. Make your entity unambiguous. Engines cross-check who you are. Your company name, what you do, your founders, and your core facts should be stated consistently across your site, LinkedIn, industry directories, and reputable third-party pages. Use Organization, Article and FAQPage schema, and connect your identities with sameAs, so the machine never has to guess which "you" it is reading.
4. Publish things the AI cannot generate on its own. Original data, first-party benchmarks, real case-study numbers, a clear and defensible opinion. A model can paraphrase commodity advice without crediting anyone. It cannot invent your proprietary dataset, so when it uses your finding it has to name you. This is the most durable form of citation insurance you can buy.
5. Structure for extraction. Descriptive headings phrased as the questions customers actually ask, short paragraphs, clean lists, a real FAQ block, and a logical information architecture. The easier you are to parse, the more often you get parsed.
6. Earn third-party authority off-site. Generative engines weigh what credible others say about you, not just what you say about yourself. A mention in a publication the model already trusts can outperform a dozen of your own posts. This is exactly where GEO and digital PR fuse, a theme we develop in digital PR as the GEO moat.

One practical discipline ties it together: track it. Run your top 20 buying-intent questions through Google's AI Overviews on a schedule and log whether you are named, what is said about you, and which sources got cited instead. That citation log is your new rank tracker, and it tells you exactly which third-party sources to go earn next.

A 30-day citation playbook

The list above is the philosophy. Here is the sequence we actually run when a client wants movement inside a quarter. It is deliberately concrete so a small team can execute it without us in the room.

1. Week 1, build the question map. Write down the 20 to 40 questions a real customer asks on the path to choosing you. Not keywords, questions. Phrase them the way a person types them into Google or speaks them to an assistant. This list is the backbone of everything that follows.
2. Week 1, run the baseline audit. Search each question in Google and record whether an Overview appears, whether you are cited, who else is cited, and what the box says. This is your starting scoreboard. Without it you cannot prove progress later.
3. Week 2, fix extraction on your best pages. Take the ten pages closest to those questions and rewrite the opening so the answer lands in the first two sentences. Add a real FAQ block. Tighten headings into question form. This alone often moves you into Overviews you were absent from.
4. Week 2, lock down your entity. Audit your name, description, and key facts across your site, LinkedIn, and major directories. Make them identical. Add or correct Organization and Article schema with sameAs links. Ambiguity here quietly costs you citations.
5. Week 3, create one thing the AI cannot. Ship a single piece of original value: a small dataset, a benchmark, a named case study with real numbers, or a genuinely opinionated framework. One is enough to start. It becomes a citation magnet because it cannot be synthesized from the open web.
6. Week 3 to 4, earn one credible mention. Pitch the original asset to a publication or expert the model already trusts. One strong third-party citation often outperforms a month of your own posts because the engine weighs outside voices more heavily than your own.
7. Week 4, re-run the scoreboard and decide. Search the same questions again. Compare against the baseline. Wherever a competitor is still cited and you are not, you now know exactly which page to deepen or which third-party source to go earn next. Repeat the loop.

Optimizing across engines, not just Google

Google's Overview is the largest surface, but it is not the only one. ChatGPT, Perplexity, Copilot, and the assistants baked into browsers and phones all compose answers and cite sources, and they do not weigh the same signals. Some lean heavily on the live web; some lean on what they were trained on plus a retrieval layer; some surface a tidy citation list and some bury attribution. The mistake is to treat "AI search" as one channel with one checklist. The durable signals (a clear answer up front, a strong entity, original data, third-party trust) help everywhere. The tactical specifics differ enough that you should measure each engine on its own and let the data, not a generic listicle, decide where the next hour of effort goes.

Measuring impact beyond raw sessions

The trap is judging this era by the one metric that is structurally guaranteed to fall: total organic sessions. If sessions are your headline number, AI Overviews will look like a catastrophe even in cases where your actual business is improving. Change what you watch.

• Conversions and revenue per session, not session count. If you shed 10,000 low-intent visits and lose almost no conversions, your traffic just got more efficient. Track conversion rate and revenue against the smaller, warmer pool that still arrives. Quality up beats volume flat.
• Assisted conversions and the full path. An AI Overview that displays your brand without a click still does work. It plants the name. The customer searches you directly a few days later, or types your URL, and converts through what looks like "direct" or "branded" traffic. Multi-touch attribution and assisted-conversion reporting catch this; last-click reporting credits it to the wrong channel and makes your best content look worthless.
• Branded search lift. One of the cleanest signals that Overviews are helping rather than only hurting: people search your brand name more. Watch impressions and clicks on branded queries in Search Console. If unbranded informational clicks fall while branded searches rise, the Overview is functioning as a billboard. You are being seen and remembered, then chosen later.
• Citation share of voice. Across your priority questions, how often are you named in the Overview versus your competitors? This is a leading indicator. It moves before revenue does, and it is the metric your content and PR efforts should be steering.
• Impressions versus clicks in Search Console. Rising impressions with falling click-through on informational terms is the fingerprint of Overview interception. Reading that gap by query type tells you precisely where you are being summarized rather than visited, and where to stop spending effort chasing clicks that will not come.

The metrics dashboard, old vs. new

If you only change one thing after reading this, change what sits at the top of your reporting. The table below contrasts the metric that will mislead you with the one that tells the truth, for each question you are likely to ask.

Put plainly: if you only measure sessions, you will make the wrong decisions. You will kill content that is quietly seeding branded demand and double down on volume that no longer converts. Measure the journey, not the doorway.

A defensive content strategy that still wins

Knowing which queries are exposed tells you where to stop pouring effort. Here is where to redirect it.

• Stop scaling thin explainers. If a page exists only to define a term or answer a one-line fact, it is now competing directly with the summary box and will keep losing. Consolidate these into deeper, genuinely authoritative resources, or retire them. Quantity of shallow pages is a liability now, not an asset.
• Move up the value curve into work the AI cannot replicate. Original research, proprietary benchmarks, frameworks with a point of view, named case studies with real numbers, expert analysis that takes a position. This content earns citations precisely because it cannot be synthesized from the open web, and the readers it pulls through are the ones who convert.
• Build for the high-intent, click-protected queries. Strong product, pricing, comparison, and decision-stage pages. This is where clicks survive and revenue lives. Many businesses over-invested in top-of-funnel volume and under-invested here; AI Overviews make rebalancing toward the bottom of the funnel the obvious move.
• Invest in brand and entity strength. The more recognized your brand, the more the Overview surfaces you, and the more "zero-click impressions" convert into direct and branded visits later. This is where GEO and digital PR pay off twice: they earn citations and build the brand recall that turns a no-click impression into a customer.
• Keep the technical foundation immaculate. Generative answers are built on the indexed web. If your pages cannot be crawled, rendered, and parsed, they cannot be cited. Crawlability, fast rendering, clean structure, and schema are now the price of admission to the Overview, not an optional nicety.

Reallocating the budget, before and after

Strategy that does not touch the budget is just a wish. The shift below is the one we most often help clients make: the same spend, pointed at work that compounds in the answer economy rather than work that competed for clicks the box now intercepts.

None of this means abandoning SEO. AI Overviews run on top of Google's index; the pages that win citations are usually the same pages that deserve to rank. What changes is the goal of the work: you are now optimizing for extraction, attribution, and high-intent capture rather than raw click volume on commodity questions.

A scenario: the SaaS team that rebalanced

Consider a mid-size B2B software company, the kind we talk to often. They had built a content engine around hundreds of "what is" and "how to" posts, and it had worked for years. When Overviews arrived, that library lost a third of its traffic within two quarters, and the internal reaction was to cut the content budget because "the channel is dead." That would have been the expensive mistake. Instead they froze new thin-post production, audited the library, kept the dozen pages that actually fed pipeline, and redirected the freed-up time into one original benchmark report and a handful of decision-stage comparison pages. Within a couple of quarters the benchmark was being cited inside Overviews and by a few assistants, branded search was climbing, and the comparison pages were converting at several times the rate of the explainers they replaced. Sessions never returned to the old peak. Pipeline from organic was higher than it had ever been. The lesson was not "do more content" or "do less." It was "do different content, and measure it differently."

A practical lens for China-to-US brands

For brands selling into the US from China, the answer layer raises the stakes in a particular way, and it is worth a section of its own. US buyers increasingly open a question to an assistant before they open a browser: "best supplier for X," "is brand Y legit," "X versus Z, which is better for a US warehouse." When the answer is composed, Chinese suppliers and 出海 brands are frequently invisible, not because the product is weak but because nobody has built the English-language, third-party trust signals the engine reads. The factory is excellent and the website is fine, but the open web has little to say about the brand in the language and on the surfaces the model trusts, so the model has nothing to cite.

This compounds a problem we have written about elsewhere. If your US presence already leans on intermediaries or marketplaces, the answer layer can quietly route demand to whoever owns the narrative, which is rarely you. We unpack the underlying dynamics in building a brand versus renting marketplace traffic and in the overseas middleman traps piece. The defensive move is the same one a US-native brand makes, with extra urgency: own a clear English-language entity, publish original proof that an engine can cite, and earn third-party mentions on surfaces the model already trusts. For B2B exporters specifically, the demand-generation mechanics layer on top of this, and we keep that playbook in the China B2B export playbook.

One honest caution on B2B execution, because it matters for how this work is scoped. When we help an exporter with prospect demand, our role is to deliver a verified, well-researched list and the positioning to use it. The client runs the outreach to their own prospects. We do not contact your customers or prospects on your behalf, and we keep the agency fee for any creator or KOL work separate from the creators' own fees. The point of saying so here is that "becoming citable" and "generating B2B demand" are related but distinct programs, and conflating them leads to disappointment on both.

The mistakes that make it worse

Most of the damage we see is self-inflicted, a reaction to the traffic dip rather than to the underlying shift. A few patterns recur often enough to flag.

• Blocking AI crawlers in a panic. The instinct to wall off your content so the AI "stops stealing" it is understandable and almost always wrong. If the engine cannot read you, it cannot cite you, and you trade a no-click impression that builds brand recall for total invisibility. There are narrow cases for restricting specific bots, but a blanket block usually amputates the very visibility you are trying to defend.
• Cutting content budget because sessions fell. The reflex to defund content when the pageview chart drops is exactly backwards. The thin pages should go, but the original, citation-worthy work is what earns you a place in the answer. Cutting it hands the citation to a competitor and removes the brand recall that was quietly feeding your direct and branded traffic.
• Chasing every lost informational keyword. Pouring effort into reclaiming clicks on definitional queries that the Overview now owns is throwing money at a structural change. That ground is gone. Redirect the budget to click-protected, high-intent pages and to the original content that gets cited.
• Optimizing only for ChatGPT or only for Google. The answer engines do not cite identically, and a tactic that helps one can be irrelevant to another. Treating "AI search" as a single channel leads to misplaced effort. Measure each engine on its own and let the citation data, not a generic checklist, decide where you invest.
• Declaring victory on impressions alone. Being shown in an Overview is necessary but not sufficient. If the impression never converts into a branded search, a direct visit, or a sale, it is a vanity signal. Tie the appearance to a downstream outcome before you call it a win.
• Stuffing content with keywords to "look citable." The old keyword-density reflex is poison here. Engines reward clear, well-attributed, genuinely useful passages, not pages engineered to trip a robot. Over-optimized copy reads as untrustworthy to the exact systems you are trying to win.
• Treating it as a one-time project. The answer layer changes month to month: which queries trigger an Overview, who gets cited, how the box is laid out. A citation you earned in the spring can be gone by the fall. This is a standing program with a recurring scoreboard, not a campaign you finish.

Your metrics-to-watch checklist

Pin this somewhere your team will see it monthly. If a number here is moving the right way, you are winning the era even if your session count is not.

• Conversions per organic session. Should hold or rise as low-intent volume leaves. The single most important line.
• Revenue or pipeline from organic. The number that actually pays for the program. Watch it monthly against sessions to keep perspective.
• Branded search impressions and clicks. Rising branded demand is the fingerprint of a no-click impression doing its job.
• Citation share of voice on your priority questions. How often you are named versus competitors. The leading indicator that moves first.
• Click-through rate by intent bucket. Informational falling while transactional holds is expected and healthy, not alarming.
• Assisted conversions and direct-traffic lift. The downstream proof that visibility without a click still feeds revenue.
• Crawlability and index coverage. A boring metric that gates everything: if you cannot be crawled, you cannot be cited.
• Count of original, citation-worthy assets shipped. An input metric, but the one that drives every output above over time.

How Ignite runs it

At Ignite Consulting we treat this as one program, not a panic response to a traffic dip. We start with a visibility audit that separates your exposed traffic from your protected traffic, then maps where you are absent inside live AI answers and where competitors are being cited in your place. From there we rebuild the foundation (technical SEO, structure, schema), reshape priority content for citation and high-intent capture, and earn the third-party authority signals that make engines trust your domain, pairing on-page work with digital PR and, where it accelerates demand, paid media. We run SEO and GEO as a single engine, and report against citations, conversions, and pipeline, not the vanity session count that this era is designed to deflate. If your growth motion is mostly B2B, we connect this to the named-account work we describe in B2B ABM for named accounts and the broader B2B growth engine; if it is consumer, we tie it into the B2C growth playbook.

Frequently asked questions