B2B marketing when AI answers the question first

For B2B marketing, AI search changes the job from winning every informational click to becoming part of the answer and creating demand elsewhere. Keep publishing useful pages, but put more effort into original evidence, product and comparison content, brand demand, and channels you control. An AI citation can shape a shortlist even when it sends no visit.

That makes the channel mix matter more. Search still captures buyers with a specific vendor or purchase question. Email still works when people want it. But a physical letter, postcard, or well-chosen package reaches a desk without an ad auction or inbox filter. Direct mail deserves a measured test budget, especially for a short list of valuable accounts. It doesn't deserve the whole budget.

What B2B marketing in AI search looks like now

Google launched AI Overviews broadly in the United States in May 2024. The change was visible at the top of the results page: for many research questions, Google could assemble an answer before the familiar links. ChatGPT, Perplexity, Gemini, and other assistants trained buyers to ask longer questions and keep researching inside a conversation.

The effect on clicks is no longer guesswork. Pew Research Center's July 2025 analysis examined 68,879 Google searches made by 900 U.S. adults in March 2025. When an AI summary appeared, people clicked a standard result on 8% of visits, compared with 15% on pages without one. They clicked a source inside the summary on only 1% of visits.

That's strong behavioral evidence, with limits. It covers Google, one month, and a panel of U.S. adults. It isn't a B2B buying study, and it doesn't prove what happens in a six-month software purchase.

A smaller B2B study fills in part of that gap. Grey Matter compared 5,159 matched queries across 18 B2B Search Console accounts from the second quarter of 2025 to the same quarter in 2026. Among queries whose ranking position stayed within two places, informational click-through rate fell 38%. Commercial queries fell 20%, while branded CTR was nearly flat, down 2%. The agency correctly calls this correlation, not proof that AI Overviews caused every lost click.

The pattern makes sense. A model can answer "what is revenue attribution?" without sending a buyer anywhere. It has a harder time completing "Acme versus Contoso for a 200-seat finance team" because the buyer wants current features, pricing, proof, and a real company to stand behind the claims. Branded and bottom-of-funnel pages still have work to do after the summary ends.

I've stopped treating raw blog sessions as the main score for educational content. Watch citations on a fixed set of buyer questions, impressions, branded search, direct visits, qualified pipeline, and what prospects say influenced the shortlist. None is a clean attribution system on its own. Together they show whether the market is learning your name.

What earns an AI citation, and what doesn't

There is no guaranteed citation recipe. Anyone selling one is selling certainty the platforms don't provide.

Start with an answer a person can lift intact. Put it near the top, state the conditions, and name the source behind any number. Then add material that can't be assembled by rewriting the current top results: your methodology, a useful dataset, product documentation, a real expert's judgment, or an honest comparison. Primary evidence gives an answer engine something worth attributing.

Technical basics still count. Keep the important copy in crawlable HTML. Use descriptive headings, internal links, stable URLs, and pages that load without making a crawler fight the application. Structured data should match visible content and can qualify a page for conventional rich results. It is not special markup for AI answers.

Google's current guidance for generative AI search, updated in July 2026, says exactly that. A page must be indexable and eligible for a search snippet. Google recommends crawlable, useful, non-commodity content. It also says Google Search ignores llms.txt, and that the file neither helps nor hurts rankings or visibility in its generative features.

So should you publish llms.txt? Only as a cheap experiment for another service that says it uses the file, or to provide a tidy map for agents. Don't call it an AI Overview tactic. The widely repeated claim that every B2B site needs llms.txt cannot be verified against Google's own documentation.

The content plan also changes. Defend product, use-case, integration, alternative, and comparison pages before commissioning another broad glossary. Refresh dates and specs. Put the answer before the company throat-clearing. For informational subjects, publish fewer pages with stronger evidence. This fits a broader shift toward B2B channels that still work, where usefulness and distribution beat sheer publishing volume.

Why physical outbound gains relative value

When a model mediates discovery, a company has less control over the route between question and website. Owned audiences become more valuable because you can reach them again. So do channels that don't require a search result click.

Digital acquisition has also become dearer in broad benchmark data. WordStream by LocaliQ analyzed more than 16,000 U.S. search-ad campaigns for its 2025 report. Average cost per lead across 23 industries rose from $66.69 in 2024 to $70.11 in 2025, after a much larger increase the prior year. That isn't a B2B rate card. It is credible evidence that auction traffic hasn't become cheaper in aggregate.

Cold email has a narrower margin for sloppiness too. Since February 2024, Google has required all senders to personal Gmail accounts to authenticate mail and keep the reported spam rate below 0.3%. Senders at roughly 5,000 messages a day face SPF, DKIM, DMARC, alignment, and one-click unsubscribe requirements. Yahoo began enforcing similar standards in 2024 and also uses a spam complaint ceiling below 0.3% for bulk senders. These are sound anti-abuse rules. They also punish weak lists and unwanted volume faster than the old inbox did.

There isn't a credible universal percentage for "B2B inbox saturation" or "ad saturation," so I won't manufacture one. Higher paid-search lead costs and stricter provider enforcement are the observable parts. Your own reply, complaint, frequency, and auction data should decide whether your audience has had enough.

Mail avoids both gates. A targeted piece can put a company name, a useful observation, and a next step in front of an executive who never clicked the educational article that introduced the category. Sales can then follow with email and a call. That's the real role, not a sentimental claim that paper always wins. See direct mail versus email as a sequencing choice, not a cage match.

There is evidence of renewed budget interest, but it needs a label. Lob's 2025 State of Direct Mail survey, run with Comperemedia, covered 405 professionals at North American companies with at least 500 employees and direct input into mail. Among its 200 marketing executives, 82% said their direct-mail investment would increase in 2025. That's a vendor-sponsored enterprise survey about plans. Meanwhile, USPS reported that total Marketing Mail volume declined 1.3% in fiscal 2025. Direct mail is rising in some large-company budgets, not sweeping the whole postal system.

Published response figures need the same restraint. The ANA Response Rate Report 2023, released in February 2024, reported a 15.6% response rate for house-file direct mail and 10.8% for prospect files. Only 26 and 25 organizations supplied those response figures, the results were self-reported, and the report blended business and consumer campaigns. I would not put 10.8% into a cold B2B forecast.

USPS's 2023 Household Diary Study found households read 39% of Marketing Mail pieces and scanned another 20%. It also warned that intended response runs higher than actual response. Useful attention evidence, yes. A B2B meeting benchmark, no. The older DMA reports that circulate in sales decks have the same mixing problem, plus age. Our B2B direct-mail response-rate guide uses practical ranges and insists on defining what "response" means before comparing numbers.

Give direct mail a hard test, not a halo

Use mail where account economics can carry it. A named-account campaign for a high-value service is a better first test than blanketing every contact in the CRM. Pick one segment where sales has a real reason to call, one offer, and one primary outcome such as qualified meetings.

Then accept the constraints. Mail costs dollars per touch rather than email's tiny marginal send cost. Copy and creative turn around in weeks, not hours. You need verified postal addresses and a plan for remote employees. A physical program cannot add a vast new audience with one click. Build and verify the B2B mailing list before spending money on a clever format.

Randomly hold back a comparable group when the list is large enough. Give mailed and unmailed accounts the same sales treatment, then compare qualified meetings, opportunities, and revenue through a fixed response window. Track QR scans and landing-page visits for diagnosis, but don't confuse them with pipeline. Matchback matters because a recipient may type the brand name later, which is exactly the behavior AI-mediated discovery makes more common.

Start small enough to learn and large enough to read. If the mailed group creates no credible lift, stop or change the audience. If it produces better opportunities twice, fund another wave. That's a channel earning its place.

Frequently asked questions

How is AI search changing B2B marketing?

It answers more research questions before the buyer visits a vendor site. That hurts informational clicks first. B2B teams still need searchable content, but the target now includes citations, accurate brand positioning, branded demand, and pipeline that may arrive later through a direct visit or sales conversation. The old traffic chart misses part of the trip.

How do I get my B2B content cited by AI search?

Give the engine a clean, supportable answer and something original to cite. First-party data, current documentation, named primary sources, and candid comparisons beat a rewritten roundup. Keep it crawlable. No magic schema, and no promise that a heading trick will force inclusion.

Does llms.txt help with Google AI Overviews?

No. Google says it ignores llms.txt for Search, including generative features. You can maintain the file for another agent or service that explicitly uses it, but I wouldn't put it ahead of fixing blocked pages, stale product facts, or thin comparison content. Those are real problems.

Is direct mail worth testing for B2B demand generation?

Yes, if a won account can repay an expensive touch and you have a short, accurate list. Don't use the ANA headline response rate as your forecast. Mail a controlled group, hold some accounts back, and judge incremental qualified pipeline. It moves slowly. That's tolerable when the test is designed before the envelopes leave.