Liz Reid Conceded AI Mode Queries Run 2-3x Longer Than Your Keyword Tool Tracks

Google's Liz Reid told Bloomberg on April 23, 2026 that AI Mode queries run two to three times longer than traditional search, and AI Overviews drove more than 10% additional global queries in Q3 2025. Queries with eight-plus words inside an AI Overview have grown 7x since May 2024. Most keyword research tools still plan against 1-2 word roots, which puts them roughly an era behind actual demand.

Reid's Bloomberg Odd Lots appearance was billed as a conversation about whether search will survive AI. The more valuable part, for anyone still planning content against a two-word keyword spreadsheet, was quieter. Reid conceded, in plain terms, that the unit of demand changed. Nobody at Google was going to say "keyword planning is obsolete" on a podcast. She said the version of it that matters.

The 7x growth curve most planners still haven't repriced against

Reid's longer-queries framing is not aesthetic. The numbers underneath it are steep. According to BrightEdge's AI Overview analysis, the share of AI Overview surfaces triggered by eight-plus-word queries has grown 7x since the May 2024 launch. That is a two-year compound shift in what earns the largest visibility slot on the SERP, and most content teams still plan against Google Keyword Planner exports that cap out at phrases like "meal prep ideas" or "best running shoes."

Reid's own framing is even more direct. In her words, users are "stopping talking in keywordese" and starting to "give the computer their actual need and expect us to do the translation." I read that as Google quietly conceding the keyword abstraction layer is dying, and the planning industry hasn't quite caught up.

What changes in practice: the keyword list is now a taxonomy of intents, not phrases. If the planner's head term is "roth ira rules," the AI Mode version looks more like "can i backdoor roth if my employer doesn't offer a 401k and i already maxed my spouse's ira." Users will type that in full and expect Google to translate it. If your page answers the head term but not the long-form phrasing, AI Overviews will cite whoever does.

The ad math behind the "relatively stable" line

Reid told Bloomberg that ad revenue under AI Overviews has been "relatively stable," explaining that lower per-query click rates get offset by higher total search volume. She is probably telling the truth at the aggregate. The problem is what aggregate truth hides from the advertiser reading their own account report.

Per an earlier ppc.land recap of Reid's public comments, AI Overviews contributed more than 10% additional queries globally in Q3 2025. Meanwhile Adthena's commercial monitoring found only 13 ad placements across 25,000 sampled AI Overview pages (roughly 0.052% frequency) as of late 2025. The surface where more queries are landing is also the surface Google has barely monetized yet.

Put those together and the picture is uncomfortable for keyword-level planning. Google's top line holds because more searches happen. The individual advertiser buying on an affected keyword sees fewer bounce clicks (Reid's phrase for the half-second visits that leave before the page finishes rendering), which is fine if analytics already discounted them. They also see fewer total clicks on the keyword, which is not fine if the plan is rate-carded on clicks. The 10% query growth Google counts is not the same 10% the advertiser reads in their click volume report.

From what I have seen, the honest read is this: Google's aggregate number seems protected by offsetting behaviors. The advertiser's keyword-level ROAS usually isn't. Q2 planning needs to separate those two lines before it looks like volume moved when the underlying unit price actually collapsed.

Two moves with a real benchmark

First, audit your keyword set against the long-form version of the same intent. Take your top 20 commercial terms. Ask ChatGPT or Gemini the same question the way a real user would phrase it in AI Mode, with the 12-to-15-word shape Reid is describing. If the model returns a paragraph that doesn't match your landing page's opening 150 words, you're not in the retrieval set for that intent. The coverage benchmark I'd target: 80% of your top 20 commercial intents should have a page whose opening paragraph answers the long-form version of the question, not just the short-form keyword.

Second, separate bounce-click traffic from engaged traffic in your GA4 reports before you evaluate the Q1 to Q2 delta. Reid's specific claim is that AI Overviews cut the half-second visits. If your AI-Overviews-exposed pages still send engaged users at the same rate, the lost traffic was never going to convert. Sessions under 10 seconds on informational pages are a reasonable proxy for bounce-click traffic, and segmenting those out usually narrows the panic window from "down 40%" to something closer to "down 8% on meaningful visits."

For the same kind of post-update diagnostic on how to separate noise from actual ranking loss, we covered the April 23 SERP volatility spike earlier this week, and the earlier pattern of small publishers losing 60% of search traffic after the March core update. Both are symptoms of the same shift Reid is describing: Google's aggregate stays intact, publishers absorb the variance.

The slop line is deflection. The keyword-length line isn't.

Reid's comment that "before AI slop, there was slop" is defense, not a new position. She is correct that low-quality content predates generative AI. The implication that ranking quality alone will continue to sort it gets harder to defend when content can be generated at roughly a 1000-to-1 ratio against editorial output. I wouldn't build a content strategy on that framing.

The keyword-length admission is the more valuable concession. On paper, saying "queries are longer" sounds like a neutral observation. In context, it reads as Google telling working marketers that the noun their planning has used for 15 years stopped being the right unit. That is the sentence worth screenshotting from the interview.

One more wrinkle worth flagging. Reid also said Google shows ads on "less than a quarter of queries," which is both a reassurance and a warning. Reassurance, because most organic intents stay organic. Warning, because the set that does monetize is narrowing, and inside that 25% window the retrieval shape is what Reid just described: long-form, conversational, translated by Google. Which means the commercial queries you actually compete for will be the ones least resembling the phrases your planner is ranked against. That is a strange transition period to run Q2 through without adjusting the inputs.

For context on how Google is already restructuring the SERP around the same logic, our read on Chrome's AI Mode side panel described the same pattern from the browser side: long-form translation inside the surface, source page as a footnote after the answer.

One slide to bring to next cycle's planning doc

Put the BrightEdge 7x growth curve next to your planner's top 50 terms. If your planner's top 50 average four words and the actual query curve is running into double-digit-word territory, you're missing a translation layer between what the team plans against and what users actually type. Fix the gap before the next content sprint, not after it. From what I've seen, teams that rebuilt their keyword taxonomy around long-form intent in late 2025 are the ones whose pages have held up cleanest through the April 2026 volatility.

The keyword isn't dead as a metric. It's dead as the unit your page targets. Reid gave the public-podcast version of that admission on Bloomberg, and the teams that heard it will be repricing their content briefs this week.

Notice Me Senpai Editorial