AI Search Is Changing B2B Discovery: What to Do About It
Your buyers' first reader is now often a machine. It reads your site, decides what you do, and puts you on the shortlist or leaves you off without telling you. The verified data, what it means for manufacturers, and a self-audit to fix it.

More and more, the first reader of your website is a machine. Before a buyer fills out a form, an AI assistant may already have read your site, decided what you do, and either put you on the shortlist or left you off it. Nobody sends you a rejection email. You just get fewer of the right calls, which is a frustrating thing to try to diagnose.
Try this before you read further
Open ChatGPT, Gemini, or Perplexity. Paste your homepage URL and ask: "Based on this page, what does this company do, and who does it help?"
For a lot of good B2B firms the answer comes back as one polished, generic paragraph. Nothing in it is wrong, nothing in it is memorable, and it could describe forty other companies. The specialty your best customers can name in one sentence gets lost along the way.
That's the whole problem in small. An AI assistant scans the words on your page and checks whether they answer the buyer's specific question. If your site gives a vague answer to a specific question, the AI passes the vague answer along, and the buyer moves on to a supplier it could describe.
Your buyers are starting with the machine
This isn't fringe behavior anymore. In G2's March 2026 survey of 1,076 B2B software buyers, 51 percent said they now start their research with an AI chatbot more often than with Google, up from 29 percent the year before, and 71 percent rely on AI chatbots at some point in their research (G2, 2026).
It shapes the decision as well as the search. In G2's April 2025 survey of 1,169 B2B decision-makers, generative AI chatbots were the most influential source for building a vendor shortlist, at 17 percent, ahead of software review sites at 15 percent and ahead of vendor websites (G2, 2025). And in the 2026 survey, one in three buyers said they had purchased from a vendor they had never heard of before (G2, 2026). That's either a threat or an opening, depending on whether an AI can find a clear, credible reason to name you.
Buyers do still want a human. Gartner found that 69 percent of B2B buyers prefer to validate AI-generated insights with sales reps, in a survey of 645 buyers (Gartner, 2026). The AI helps build the list and people still close the deal, so getting named is now necessary but has never been enough on its own.
Meanwhile, the click you used to earn is shrinking
Even when you rank, the visit is worth less. When Google shows an AI summary, people click a traditional link on 8 percent of those searches, compared with 15 percent when there is no summary, and they click a link inside the summary itself just 1 percent of the time (Pew Research Center, 2025).
Ahrefs measured AI Overviews cutting clicks to the top-ranking page by 34.5 percent in April 2025, and its December 2025 update put the drop in click-through rate for position one at 58 percent (Ahrefs, 2026). In the first four months of 2026, 68 percent of US Google searches ended without a click (SparkToro, 2026).
Ranking first matters less when a machine-written answer sits above you and settles the question on the spot. More of the value now goes to the sites the answer is built from.
What this looks like for manufacturers
Most of the survey data above comes from software buyers. Industrial buying is slower to change, and the numbers say so. In the 2026 State of Marketing to Engineers survey from TREW Marketing and GlobalSpec, 63 percent of engineers and technical professionals said they rarely or never use generative AI to help evaluate or make a work-related purchase (31 percent never, down from 42 percent in 2025, and 32 percent rarely; n = 751), and their trust in AI answers averaged 4.7 out of 10, up from 4.4 in 2025 but still below 6.5 in 2024 (TREW Marketing and GlobalSpec, 2026). In the 2025 edition, ChatGPT was the most-used tool among those who do use AI, at 72 percent (TREW Marketing, 2025).
So for a manufacturer, AI is still an early channel rather than the main one. But the buyers who do use it ask very specific questions, and they ask them early. A typical prompt looks something like this:
- "Find US contract manufacturers that do 5-axis CNC machining of titanium for medical devices, ISO 13485 certified, for runs under 5,000 parts."
- "Which powder coating suppliers in the Midwest can handle parts up to 20 feet long?"
- "Compare three suppliers of food-grade stainless conveyor systems with short lead times."
Read those as a checklist. The AI can only name you if your site says, in plain text it can read, what process you run, on what materials, to what tolerances and sizes, under which certifications, for which industries, from where, and at what volumes and lead times. That's the same list an engineer checks by hand, and it's why the work that gets you named by an AI is the same work that wins the human review. We walk through that review in what manufacturing buyers check on your website. Peter, our co-founder, calls this the 20-second test.
Here's what a supplier site needs for this:
- One page per process or product family. "Capabilities" as a single page with a bulleted list gives an AI nothing to match a specific question against. A page for CNC machining, one for sheet metal, one for finishing, each with its own specs, does.
- Specs in text, not trapped in PDFs or images. Tolerances, materials, part sizes, volumes, and equipment lists belong on the page. Keep the downloadable datasheet too, but don't make it the only place the numbers live. Google's own guidance for its AI features stresses keeping important content in text form (Google Search Central).
- Certifications stated with their scope. "ISO 9001:2015 certified for the design and manufacture of..." beats a row of logos, which a machine can't read and a buyer can't verify.
- Industries and locations named. If you serve aerospace and medical from plants in Ohio and Mexico, say so in words on the relevant pages.
- Nothing blocking the crawlers. Google says there are no special requirements or markup needed to appear in its AI features, beyond being indexed and eligible for a snippet, and that structured data should match the visible text (Google Search Central). OpenAI says sites that want to appear in ChatGPT search should allow its OAI-SearchBot crawler in robots.txt (OpenAI). Check that your robots.txt, CDN or firewall isn't blocking either one without you knowing.
For a complex catalog, structure matters as much as copy. When we rebuilt the site for Telling Industries, a manufacturer with a technically complex product line, we built a custom product configurator so buyers could find the right configuration themselves, and qualified leads tripled after launch. A catalog a buyer can find their way through is usually one a machine can make sense of too.
This is a positioning problem, not a plugin problem
The reflex is to treat this as a technical chore. Bolt on some FAQ schema, add a few question-shaped headings, call it "answer engine optimization," and move on. That misreads the shift, because schema can't fix vague copy. It can help a machine read a page whose words are already clear, and it does nothing for a page whose words are wrong.
Ranking got you found. Citability gets you named.
What changed is where the decision forms. Buyers build a shortlist, and often a preference, before they talk to you, and sometimes before they land on your site. So "how do we rank" is only part of the question now. The bigger part is whether, when an AI or a person decides who belongs on the list, there's a clear, consistent, believable case to point to. That's a positioning and content problem first, and a technical one second.
There's research behind this. In the peer-reviewed paper that named the field, the best-performing methods, which included adding quotations, statistics, and citations to credible sources, improved a page's visibility in AI-generated answers by up to about 40 percent, while old-school keyword stuffing performed worse than doing nothing (Aggarwal et al., "GEO: Generative Engine Optimization," KDD 2024). And Pew found that 88 percent of Google's AI summaries cite three or more sources, so the goal is to be one of the sources the machine trusts enough to name.
What to actually do about it
None of these moves are new, but AI search has raised the price of ignoring them. Our docs guide on writing for AI answers shows how to shape a page so an assistant can quote it.
Run the 20-second test on your own site
Paste your homepage URL into a chatbot and ask what your company does and who it's for. Then ask it the kind of question a buyer would: "Which suppliers can [your process] for [your industry]?" Read both answers the way a buyer would, with 11 tabs open and 90 seconds to decide. Whatever separates what the AI says from what your best customers say is the work in front of you.
Make one clear promise per page, in the buyer's words
A page that tries to say three things gets summarized as none of them. Give each page one plain claim that names your category and your buyer, in the language your customers use on sales calls rather than the language of your org chart. Where three overlapping pages say similar things, collapse them into one strong page.
Put your proof in text, not trapped in images
The facts that make you credible, the specs, certifications, customer outcomes, and quotes, have to exist as readable text an AI can lift. A machine can't see a logo wall or a testimonial baked into a graphic. Original numbers from your own work beat borrowed industry stats, because no competitor can copy them.
Earn corroboration off your own site
An answer engine trusts a claim more when someone other than you makes it. Listings on the supplier directories and industry associations your buyers use, a trade publication article, a conference talk, or a customer case study on their site all let an AI confirm your positioning against a second source. Keep the story consistent across your site, LinkedIn, and listings.
Keep it current, and measure the new channel
Date your content and keep specs, certifications, and equipment lists up to date. Then measure: add AI referral sources like ChatGPT, Perplexity, and Gemini as their own segment in your analytics, and watch how many of those visitors request a quote. Google counts traffic from its AI features in the regular Search Console performance report, so you won't see it separately there.
Now for the part this whole article has been leading to. Score yourself.
Can an AI recommend you?
Score your own firm. Tick each one you can honestly answer yes, then count your checks against the guide below. Do the first one for real: open a chatbot, paste your homepage URL, and ask what your company does and who it is for.
- 5 to 6
- an AI can already sell for you.
- 3 to 4
- a coin flip. Fixable, but leaking.
- 0 to 2
- you are being quietly skipped.
The honest part
We're not AI skeptics. We use these tools every day, and when we say buyers lean on them, we're describing a pull we feel ourselves.
We're also not selling panic. Some of the biggest numbers in this space are contested. Gartner's forecast that search engine volume would drop 25 percent by 2026 was a prediction rather than a measurement (Gartner, 2024). Google disputes the Pew study's methodology. The original GEO research tested much of its method on a simulated engine rather than today's live ChatGPT or Gemini. From January to April 2026, only 0.34 percent of Google searches went to AI Mode (SparkToro, 2026), though Google said at I/O 2026 that AI Mode use was growing fast, so expect that share to rise. And most engineers still rarely use AI to buy, as the TREW data shows.
The forecast doesn't need to be exactly right, only right about the direction, and the independent sources agree on the direction. That doesn't mean abandoning search for a new acronym. Do both: keep earning rankings through solid SEO and content, and build the brand and clear positioning that make you the obvious answer when the decision happens somewhere you can't see. When an AI does send you a visitor, make that click count with a page built to convert it, which is the job of our Conversion System.
That version of SEO holds up whichever way the numbers land.
Keep reading
Branding & Digital MarketingWhere Your Quote Requests Stall: A 12-Question Manufacturing Lead Generation Check
Most manufacturing lead generation advice is a list of ways to get more traffic. Before you buy more, find out where the interest you already have stops. Here are twelve questions that trace a quote request from the first click to the estimator's desk, and what to fix at each one.
Branding & Digital MarketingHave You Outgrown Your Brand? Refresh, Rebrand, or Fix the Website
Most established B2B companies that feel they've outgrown their brand have outgrown their website and their message. The name and the logo are usually fine. Here are the signs, what each one points to, a self-assessment you can run with your team, and a decision rule for refresh, rebrand or leave it.
Branding & Digital MarketingBrand Is a B2B Growth Lever, Not a Paint Job
Brand gets filed under the soft stuff right up until you see the numbers. Here's the business case for brand in B2B and manufacturing, and how to run it like the revenue engine it is.
Ready to talk about your site?
Book a call with Chykalophia. We focus on your business, your goals and what is in the way, and tell you fast if we are a fit.
Strategies for businesses building what's next.
Occasional notes on turning B2B websites into pipeline. Unsubscribe any time.