Google AI Overviews Are Unreliable: A Marketer’s Action Plan

Google’s AI-powered search summaries just displayed chatbot-style output in response to a simple, single-word query — and if you’re still treating AI Overviews as a stable, predictable channel, this incident should fundamentally change your thinking. The “disregard” bug, spotted on May 22, 2026 and reported by The Verge, exposed a fragility in how Google’s AI search layer processes user queries — one with direct consequences for how your content surfaces, gets summarized, and ultimately drives or kills clicks. This isn’t a one-off glitch. It’s a signal about the structural instability baked into the AI search era.

What Happened

On May 22, 2026, users who searched for the word “disregard” in Google encountered something that had no place in a search results page: instead of the standard AI Overview panel — the condensed, AI-generated answer that Google has been placing at the top of results since 2024 — they got a response that looked like output from a conversational AI chatbot. Not a summary of web content. Not a digest of relevant articles. Something that resembled what you’d see if you typed a prompt directly into an AI assistant and the model went off-script.

According to The Verge, the behavior was spotted on X (formerly Twitter) and quickly documented in screenshots. The Verge described the AI Overview response as looking “like what you’d see from a more traditional AI chatbot instead of the typical AI summary.” The article was published May 22, 2026 and the bug appeared to affect the AI Overview panel specifically for searches containing the word “disregard.” Note: The Verge’s article was inaccessible for full retrieval at time of writing; this post draws on the article’s title, URL, publication date, and the topic summary on record.

To understand why this matters, you need to understand what the word “disregard” means to a language model. In the world of AI prompt engineering and security research, “disregard” is far from a neutral term. It’s one of the most commonly used words in prompt injection attacks — deliberate attempts to manipulate an AI system into ignoring its instructions or constraints by inserting adversarial text. Phrases like “disregard your previous instructions,” “disregard the above,” and “disregard your system prompt” appear in virtually every documented library of prompt injection techniques. These phrases exploit how language models process text sequentially: if an input can be framed as an instruction overriding the model’s prior context, the model may comply.

What the “disregard” bug suggests is that Google’s AI Overviews system — somewhere in its query-to-summary pipeline — confused the search query itself with a command to the underlying language model. The word “disregard,” entered by a user with zero adversarial intent, may have been enough to trigger the model’s sensitivity to that vocabulary. Rather than executing its search summarization function, the model responded in a generative, chatbot-style mode — a mode that should not be reachable from a standard Google search interface.

This is what security researchers have described as a prompt injection vulnerability manifesting in a production system, at scale, triggered not by a sophisticated attacker crafting an exploit, but by an ordinary English word in an ordinary search. That distinction matters enormously: this wasn’t a hacker probing the system. This was a user asking Google something and getting back something the system was never designed to show.

For context: Google AI Overviews launched broadly at Google I/O in May 2024. The product was previously called Search Generative Experience (SGE) during its extended testing period. The public launch did not go smoothly. Within days of broad availability, researchers and users compiled a growing catalogue of AI Overview failures: the system recommended users eat rocks for nutritional benefits after apparently synthesizing content from a satirical source; it suggested adding glue to pizza sauce to help cheese stick, having cited a Reddit comment that was made as a joke; and it generated other factually absurd summaries that spread widely on social media. Google moved quickly to remove the most egregious examples and implemented additional source quality filtering and content safeguards, but the fundamental reliability question was never fully closed.

The “disregard” incident represents a different category of failure than the 2024 hallucinations. Those earlier bugs were accuracy failures — the AI summarized content incorrectly or drew from low-quality sources, but it was still trying to function as a search summarizer. The 2026 “disregard” bug appears to be a behavioral failure — the AI broke out of its search summarization role entirely and responded in a mode that should not be reachable via a standard query. That is a qualitatively different problem, and it carries different implications for how marketers should assess and manage this channel.

At the time of writing, Google had not issued a public statement explaining the root cause of the bug or confirming a fix. The malfunction was documented on X in real time before being reported by The Verge, following the same pattern as the 2024 failures — public social media documentation before any official acknowledgment.

Why This Matters

If you’ve been running SEO as a core channel since 2024, you’ve been living in the AI Overviews era long enough to have adapted — or tried to. You’ve watched click-through rates shift as AI-generated summaries answer questions directly on the results page, eliminating the need for users to click through to your content. You’ve adjusted content strategy to structure information clearly, use schema markup, and build topical authority — all in service of appearing in those AI summaries, the 2025 analog of ranking in position one. The “disregard” bug forces a harder and more uncomfortable question: what happens to your search strategy when the channel you’ve been optimizing for is structurally unpredictable at a fundamental level?

The reliability problem is a brand representation problem. AI Overviews don’t just affect click volume — they affect how your brand, your products, and your expertise are described to users before they ever reach your website. If a user searches for your brand name, your product category, or a query you rank strongly for, and Google’s AI Overview generates a summary that is wrong, outdated, or — as in the “disregard” case — returns something that doesn’t resemble a legitimate search result at all, that’s a brand exposure you did not agree to and cannot directly control. You can influence what the AI summarizes through the quality and structure of your content, but you cannot control its behavior.

SEO investment assumptions break when the summarizer breaks. The dominant SEO theory for the AI era is built around a reasonable premise: structure your content for AI citability. Provide clear, factual, well-organized information. Use proper schema markup. Build E-E-A-T signals (experience, expertise, authoritativeness, trustworthiness). That theory is reasonable when the AI is operating as designed. When it malfunctions — as it demonstrably has, multiple times, across different failure modes — none of that optimization work can prevent a broken outcome. The AI might cite your content in a malfunctioning summary, ignore it entirely, or return something that has nothing to do with the original query. None of those scenarios are within your control, regardless of how well-optimized your pages are.

The affected audience scale amplifies every malfunction. Google processes an estimated 8.5 billion searches per day, according to Internet Live Stats. At Google I/O 2024, Google reported that AI Overviews were already being served to more than one billion users globally. Any instability in that system — even a bug that affects a narrow slice of queries for a few hours — reaches an audience that eclipses every other single marketing channel. The “disregard” bug was visible on X within hours of its appearance. That is the speed at which AI Overview failures can surface publicly and spread — and that is the speed at which brand damage can accumulate before anyone on your team is even aware there’s a problem.

Agencies and in-house SEO teams face a reporting blind spot. You’re being asked to show ROI from organic search in an environment where the primary interface layer can malfunction in ways that standard analytics tools won’t catch. Rank trackers report position. Google Search Console reports clicks and impressions. Neither data stream will tell you that Google’s AI Overview was displaying chatbot-style content this morning on a query your highest-value page ranks for. You’ll see the anomaly — a CTR dip, a traffic drop on a specific date — but connecting it to an AI Overview malfunction requires active monitoring that most teams simply aren’t doing.

High-stakes verticals face compounded risk. Health, finance, legal, and regulated e-commerce brands are particularly exposed. The 2024 AI Overview failures that drew the most significant public and regulatory attention were concentrated in the health category — the “eat rocks” recommendation being the most widely cited example. If a behavioral malfunction similar to the “disregard” bug were to occur on a query like “disregard [medication name] side effects” or on any query in a sensitive regulated vertical, the consequences extend well beyond a strange search result. They could include AI-generated misinformation about medical products, financial instruments, or regulated services, served to millions of users before any correction is possible.

Solopreneurs and small marketing operations are structurally disadvantaged. Large brands with dedicated search teams and enterprise monitoring tools can build systems to watch AI Overview behavior across hundreds of queries. A solopreneur, a local business, or a three-person agency does not have that infrastructure or bandwidth. The “disregard” bug reveals an asymmetric risk landscape: the marketing organizations least equipped to monitor AI Overview behavior are also the ones most likely to be blindsided when it misfires on their critical queries — and least equipped to respond effectively when it does.

The Data

The “disregard” incident fits a documented pattern of AI Overviews reliability events that stretches back to the product’s launch in May 2024. Understanding that pattern helps calibrate the actual risk level marketers are managing:

Sources: The Verge (May 2026); The Guardian, “Google’s AI Overview errors and hallucinations,” May 2024 (article inaccessible at time of writing — title and date on record); Search Engine Land, AI Overviews launch coverage, May 2024 (article inaccessible at time of writing — title and date on record).

Beyond individual malfunction events, the structural impact of AI Overviews on organic search behavior has been directionally consistent across industry analyses since 2024. Informational queries — “what is X,” “how does X work,” “explain X” — are both the query type most likely to trigger AI Overviews and the query type where AI Overviews most dramatically reduce click-through rates, because users receive a synthesized answer directly on the results page without needing to visit a website.

The final row is new as of May 2026. Before the “disregard” incident, no marketer had reason to include “queries containing prompt-injection vocabulary” as a separate risk category in their search strategy. That category now exists, and the vocabulary it covers — “disregard,” “ignore,” “override,” “pretend,” “forget,” and dozens of related terms used in AI prompt manipulation — is broader than most marketers would initially expect when reviewing their keyword portfolios.

Real-World Use Cases

Use Case 1: E-Commerce Brand Establishing AI Overview Monitoring for Product Queries

Scenario: A mid-size outdoor gear brand sells products that appear in informational queries — gear comparisons, buying guides, how-to usage questions — that regularly trigger AI Overviews. After the “disregard” bug makes headlines, the marketing manager wants to know what Google is currently saying about their products and whether AI Overviews on their key queries are behaving normally.

Implementation: Compile a list of 30–50 priority queries covering brand name, top product categories, comparison terms, and the informational queries the brand ranks strongly for. Run each query weekly in a fresh incognito browser session and screenshot the AI Overview panel when it appears. Log what each summary says, whether it cites the brand’s website, whether the information is accurate, and whether the summary looks structurally normal. Flag anything anomalous — wrong pricing, outdated product details, or any result that doesn’t look like a standard AI Overview. Route flags to the SEO lead for documentation and potential escalation to Google’s AI Overview feedback channel. This workflow can be started manually within a day and scaled with browser automation or SERP tracking tool APIs once the baseline is established.

Expected Outcome: Within 30 days, the team has a documented baseline of AI Overview behavior across their key queries and a historical record of summary content. They can catch behavioral malfunctions — including prompt-injection-adjacent anomalies — within a week of occurrence rather than discovering them from a customer complaint or a viral social media post. This level of monitoring is currently uncommon enough that it represents a competitive advantage in most verticals.

Use Case 2: B2B SaaS Content Team Stress-Testing Their AI Overview Strategy

Scenario: A content marketing team at a B2B SaaS company spent 2025 building a library of AI-Overview-optimized content: concise, well-sourced, FAQ-formatted pages designed to be cited in search summaries. The “disregard” bug triggers an internal audit of how exposed their content program is to AI Overview instability.

Implementation: Start with a channel attribution analysis — pull Google Search Console data and tag queries by AI Overview appearance frequency using a SERP features tool. Calculate what percentage of total organic traffic comes from queries where AI Overviews regularly appear. If that figure exceeds 35–40%, the content program has meaningful concentration risk. Launch a parallel content track focused on owned-channel conversion: long-form guides that drive email signups, original research reports that generate direct downloads, podcast and video content that builds a direct audience relationship. Continue producing AI-Overview-optimized content — it still has SEO value when the AI is functioning correctly — but allocate explicit budget to the owned-channel track as a portfolio hedge. Set a 12-month target to reduce AI-Overview-dependent traffic to under 30% of total organic.

Expected Outcome: Over 12 months, a dual-track content strategy reduces structural exposure to AI Overview instability while maintaining the SEO benefits of well-organized, citable content. The team has an evidence-based answer when leadership asks whether the AI Overview investment is justified: yes, and here’s how the portfolio is diversified to limit downside risk.

Use Case 3: Digital Marketing Agency Adding AI Overview Reliability to Client Reporting

Scenario: A performance marketing agency manages SEO for 20 clients across healthcare, financial services, and retail. Monthly reporting has grown more complex as clients notice traffic fluctuations that don’t align with ranking changes or site updates. Several clients are asking pointed questions about unexplained variance.

Implementation: Add an AI Overview behavior module to the standard monthly reporting template. For each client, identify the 10 highest-priority queries by traffic and conversion value. Track AI Overview appearance frequency on those queries using SERP feature monitoring available through Semrush or Ahrefs — both platforms have expanded AI Overview tracking since 2024. When traffic anomalies appear in the data, cross-reference with AI Overview behavior changes on the affected queries for that date range. Document the correlation in the client report. Include a standing contextual section explaining that AI Overview content and behavior can change independent of page rankings or site changes — and that this is now a standard search channel variable the agency actively monitors on the client’s behalf.

Expected Outcome: Clients receive reporting context that accurately reflects how search traffic is actually generated in 2026 — not through rankings alone, but through an AI summarization layer that no agency can directly control but can actively track. Churn risk from unexplained traffic variance decreases. The agency differentiates itself as more analytically rigorous than competitors still reporting only rank and sessions without accounting for the AI layer.

Use Case 4: Health Brand Building an AI Overview Compliance Monitoring Protocol

Scenario: A consumer health brand sells FDA-regulated supplements. Their products are referenced in health-related queries that commonly trigger AI Overviews, and the legal team — citing the 2024 “eat rocks” incident and the 2026 “disregard” bug — has opened a formal risk assessment of what would happen if Google generated an AI summary containing inaccurate health claims about their products.

Implementation: Cross-functional project involving marketing, legal, and regulatory affairs. Step one: document all FDA-compliant health claims that appear on product pages and in any public-facing brand content. Step two: set up weekly monitoring for queries containing the brand name plus health-related modifiers — benefits, side effects, dosage, interactions, warnings. Step three: create a documented escalation process. If an AI Overview appears with health content that contradicts compliant labeling or contains medically inaccurate information, the protocol is: screenshot immediately, submit through Google’s AI Overview feedback mechanism, notify legal, and log the incident with precise timestamps. Step four: add a visible disclaimer to product pages noting that AI-generated search summaries may not accurately reflect product information, and that users should consult the product label and a healthcare professional.

Expected Outcome: Reduced legal exposure through a documented monitoring and response trail. If an AI Overview malfunction generates harmful health misinformation about the brand’s products, the company has timestamped evidence of proactive monitoring and prompt reporting — directly relevant if regulatory scrutiny or consumer complaints follow the incident.

Use Case 5: Growth Marketer Using AI Overview Instability as a Paid Search Budget Signal

Scenario: A DTC brand’s growth marketer oversees both paid and organic search. She has observed that queries where AI Overviews appear frequently show lower organic CTR — users get what they need from the AI summary and don’t click through. She’s also observing that Google Ads CPCs on those same queries have been elevated, suggesting competitors are bidding up to capture users who skip organic results entirely.

Implementation: Build a query-level performance model mapping three variables: AI Overview appearance frequency (from a SERP features tracking tool), organic CTR (from Google Search Console), and paid CPC (from Google Ads). Queries with high AI Overview appearance frequency, suppressed organic CTR, and moderate-to-high commercial intent are strong candidates for increased paid search investment — these are queries where users who aren’t satisfied by the AI summary will scroll down to ads. Conversely, queries where AI Overviews appear inconsistently or have shown anomalous behavior (as the “disregard” incident demonstrates is possible) should be flagged for monitoring before scaling paid investment on adjacent terms. Present this framework to leadership as a dynamic budget allocation model that explicitly accounts for AI Overview behavior as a real-time variable — not a background constant.

Expected Outcome: More efficient paid/organic budget allocation within 90 days as the model identifies where paid investment has the highest expected return given AI Overview-suppressed organic performance. The framework also creates an operational discipline of treating AI Overview behavior as a live business variable, positioning the team to respond faster when the next malfunction affects queries in their portfolio.

The Bigger Picture

The “disregard” bug is a single incident, but it reveals something structural about how Google has architected its current search product — and what that architecture means for every marketing team that depends on Google Search as a channel.

For most of Google’s history, search was a deterministic system. Submit a query, retrieve a ranked list of documents. The ranking algorithm changed — sometimes dramatically, with major updates like Panda, Penguin, and Helpful Content reshaping organic rankings and causing significant traffic disruptions — but the fundamental behavior was rule-based and consistent enough that SEO as a discipline could build reliable practices around it. If you understood the signals Google weighted, you could optimize for them, measure results, and iterate with confidence.

AI Overviews change that architecture at a foundational level. They insert a generative layer — a large language model — between the user’s query and the content returned from the web. That model synthesizes, interprets, and presents information in natural language. Its output is probabilistic, not deterministic. For a given query, the AI may produce slightly different responses at different times, draw from different sources, and weight different signals — and, as the “disregard” bug demonstrates, it can be triggered into behaviors that have nothing to do with search summarization, not by a sophisticated attacker but by an ordinary English word that carries strong associations in the model’s training data.

Prompt injection as a vulnerability class is not new to the AI security research community. Researchers have documented it extensively since large language models became widely deployed in 2022 and 2023. The concept covers both direct prompt injection — where a user crafts an input designed to override model behavior — and indirect prompt injection — where malicious instructions are embedded in content that an AI retrieves from the web and then processes. What makes the “disregard” incident notable is that it appears to have been triggered by neither a malicious user nor adversarially crafted web content: it was triggered by a single common word used as a search query. That suggests the sensitivity is built into how the model processes its inputs at a more fundamental level than most people outside AI security research have appreciated.

This places Google in an uncomfortable strategic position. The company has made AI Overviews the flagship feature of its search product — the primary mechanism by which it demonstrates competitive parity with dedicated AI assistants like ChatGPT and Perplexity. But the reliability record — 2024 hallucinations, 2026 behavioral breakdown — reveals that deploying a generative language model as the primary interface for the world’s most-used information retrieval system introduces failure modes that are qualitatively different from anything in the history of search algorithm updates. Algorithm bugs can be patched with deterministic fixes. Language model behavioral anomalies are harder to reproduce, predict, and eliminate because they emerge from the probabilistic nature of the underlying models themselves.

For the marketing industry, the historical parallel that resonates most is the Facebook organic reach collapse of 2014–2016. Brands that had built their entire audience acquisition strategy on Facebook organic reach were severely damaged when the algorithm reduced that reach sharply. Brands that had diversified — that maintained email lists, built direct audience relationships, and used social media as a supplement rather than a foundation — absorbed the disruption. AI Overviews carry a structurally similar concentration risk: enormous audience access through a channel that can change in ways completely outside any marketer’s control. The “disregard” bug adds an additional dimension the Facebook comparison doesn’t cover: the channel can also break in ways that are strange, fast-moving, and genuinely difficult to anticipate.

What Smart Marketers Should Do Now

1. Set up AI Overview monitoring for your 50 highest-priority queries within the next 30 days.

If you don’t currently know what Google’s AI Overviews are saying about your brand, products, or category — in real time, with a documented history — you have a monitoring gap. Start simple: a spreadsheet, weekly manual checks in incognito browser sessions, screenshots when AI Overviews appear. Scale from there using SERP feature tracking tools. The goal in the first 30 days is baseline awareness: which key queries trigger AI Overviews, what do those summaries currently say, are they accurate, and do they look structurally normal? The “disregard” bug demonstrates that behavioral anomalies can be caught fast when monitoring is in place — and missed entirely when it isn’t, only surfacing after viral spread on social media.

2. Audit your keyword targets and ranking pages for prompt-injection-adjacent vocabulary.

The practical version of this audit is straightforward. Pull your top 100 target keywords and your highest-traffic organic queries from Google Search Console. Scan for words that carry known sensitivity in large language models: “disregard,” “ignore,” “override,” “forget,” “pretend,” “instructions,” “system,” and related terms. If any of those words appear in your target query set, you now have direct evidence — from the May 22, 2026 incident — that AI Overviews can behave anomalously on queries containing them. Add those queries to your monitoring priority list immediately, and consider adjusting content strategy around those terms until Google demonstrates a verifiable fix.

3. Reduce AI Overview query concentration in your organic traffic mix.

Run a channel attribution analysis using Google Search Console segmented by AI Overview appearance rate (available through SERP tracking tools). If more than 30–40% of your organic traffic comes from queries where AI Overviews frequently appear, you have meaningful concentration in the most volatile segment of the search channel. The diversification play is not abandoning SEO — it is funding owned-channel content in parallel: email newsletters, gated research, community platforms, and podcast content that build audience relationships AI Overview behavior cannot disrupt. The investment is real, but so is the documented risk of over-reliance on a channel where the AI can malfunction on an ordinary word.

4. Upgrade your reporting framework to make AI Overview behavior a visible, managed variable.

Monthly reports showing only rankings, sessions, and conversions no longer tell the complete story of organic search performance in 2026. Add AI Overview appearance rate on key queries, a summary of current AI Overview content for your most important branded and category queries, and a flag for any anomalous behavior observed during the reporting period. When AI Overview instability causes CTR drops or traffic anomalies, this reporting layer enables accurate explanation rather than presenting unexplained variance to stakeholders who may reasonably assume the problem is on your end. A documented AI Overview behavior history also provides valuable evidence if you ever need to demonstrate a sustained pattern of impact over time — for client presentations, for budget discussions, or for formal escalation.

5. Build a systematic feedback workflow for AI Overview errors and behavioral malfunctions.

Google’s AI Overview feedback mechanism — available directly from the AI Overview panel — is the primary escalation path for AI Overview problems that fall short of a formal legal matter. Make it part of your monitoring workflow. When an AI Overview on a key query shows inaccurate information about your brand or products, or when it displays anomalous behavior like the chatbot-style output triggered by “disregard,” submit a feedback report immediately and log that you did so with a timestamp, a screenshot, and a description of what was observed. This is not purely reactive — a documented history of proactive reporting is relevant context if AI Overview errors escalate into a PR problem, a regulatory inquiry, or a situation requiring formal legal response. It also signals to Google that sophisticated marketing teams are watching and documenting AI Overview behavior, which may influence how they prioritize fixes.

What to Watch Next

Google’s official explanation of the “disregard” incident. At the time of writing, Google had not publicly addressed the root cause of the AI Overview malfunction. If Google does explain what happened — through a public statement, a Google Search Central blog post, or a technical disclosure — that explanation will be among the most informative public disclosures about how AI Overviews process queries internally. The presence or absence of an explanation is itself a data point: transparency suggests an isolated, patchable bug; silence often signals something harder to fix.

Frequency of similar prompt-injection-adjacent anomalies across AI search. The “disregard” bug almost certainly isn’t the only query that can trigger behavioral anomalies in AI Overviews. As the product expands in language support, query type coverage, and geographic market reach throughout 2026, the surface area for unusual behavior grows. Monitor SEO communities on Reddit and LinkedIn, X, and publications like Search Engine Land and The Verge for reports of similar incidents. A cluster of related anomalies would signal a systemic architectural problem; isolated incidents suggest edge cases Google can patch individually.

Google AI Mode rollout trajectory in 2026. Google has been testing “AI Mode” — a fully conversational search experience that substantially replaces the traditional results page with an AI-driven dialogue interface. If AI Mode expands broadly in the second half of 2026, every reliability question raised by the “disregard” bug applies at higher stakes: a fully conversational AI search interface that can exhibit prompt injection behavior is a much larger problem than an AI Overview panel that behaves oddly on one query. Watch Google I/O 2026 announcements and Google Search Central communications for rollout timelines.

Competitive AI search reliability records. Microsoft Bing with Copilot integration, Perplexity, and other AI-powered search products are operating in the same generative AI search space. If Google’s reliability issues accumulate into a visible consumer trust problem, watch whether enterprise search procurement decisions or consumer behavior begin shifting toward alternatives — and whether brands start recommending alternatives for sensitive or high-stakes queries. Any meaningful shift represents both a threat (for brands heavily invested in Google AI Overview optimization) and an opportunity (to establish presence on alternative AI search channels before the competitive landscape hardens).

Regulatory attention on AI-generated search content. The European Union’s AI Act, which entered into force in stages beginning in 2024, includes provisions relevant to AI-generated content in high-stakes categories including health, finance, and employment. If AI Overviews continue to generate harmful or inaccurate content — or if behavioral malfunctions produce results that constitute actionable misinformation in regulated categories — expect regulatory scrutiny, particularly in the EU where Google already faces active antitrust enforcement related to its search dominance. In the US, the FTC’s increasing attention to AI-generated content accuracy is worth monitoring. Either regulatory environment could force structural changes to AI Overviews that materially affect every team using search as a channel.

Bottom Line

Google’s AI Overviews are a powerful product in theory and an unstable platform in practice. The “disregard” bug — documented by The Verge on May 22, 2026 — adds a new and qualitatively different failure mode to a two-year track record of AI Overview reliability problems: not just inaccuracy in what the AI summarizes, but a behavioral breakdown triggered by a single ordinary English word that caused the AI to exit its search summarization role entirely. For marketing teams, this is not an academic concern about AI limitations — it directly affects how your content is surfaced, how your brand is described to users before they reach your site, and whether the SEO investments you’re making can deliver predictable results in an environment where the summarizer itself can malfunction. The right response is not panic, and it is not abandoning search as a channel. It is proactive monitoring, deliberate channel diversification, and a clear-eyed acknowledgment that AI Overviews are a dependency with documented and ongoing reliability risk. Build your search strategy around that reality — because the teams that do it now will be substantially better positioned than those who wait for the next incident to force the conversation.