AI search is now part of B2B consideration because buyers ask answer engines to compare vendors before visiting their websites. A brand named in those answers gets a chance to be evaluated; an absent brand may never enter the conversation. CMOs should manage that presence alongside organic search, with a repeatable audit and evidence buyers can verify.
The shortlist can form before the website visit
In Forrester's State of Business Buying 2026, 94% of business buyers use AI during their buying process, based on a survey of nearly 18,000 buyers. That figure covers AI use broadly. It does not mean every buyer delegates vendor selection to ChatGPT. Forrester also describes buyers checking AI output with trusted people and external sources. Being named opens a door; credible evidence still has to support the recommendation.
Edelman, November 2026, reports that 84% of B2B decision-makers used AI to summarize a vendor's thought leadership during evaluation. Your article may reach the buying group as a paragraph inside someone else's answer, stripped of its design and conversion path. The argument has to survive that trip. A vague claim about transformation gives the buyer little to evaluate.
Start with a concrete question: when someone asks ChatGPT, Gemini, Perplexity, Claude or Google AI Overviews which firms can solve their problem, does your company appear with an accurate explanation? Search rankings remain useful, but a ranking report cannot answer that question. Neither can counting visits from AI referrals, because a buyer can read a recommendation without clicking.
Build a baseline from real buyer questions
Choose 20 to 30 questions from sales conversations, lost-deal notes and customer interviews. Keep the buyer's constraints. 'Best cybersecurity vendor' is less useful than 'Which cybersecurity providers can support a regional bank with a small internal security team?' Include comparison questions and objections raised during evaluation. Do not put your company name into every prompt. That tests recognition after prompting, rather than unaided consideration.
Ask the same questions across the five interfaces. Use fresh conversations and record the date, engine, visible model or search mode, language and location. Save the actual answer. Google may return no AI Overview for a query; record that outcome instead of quietly replacing the question. Keep branded comparison prompts separate from unbranded shortlist prompts so you do not mix two different tests.
A spreadsheet is enough. For each response, record which firms are named, whether they are recommended or merely mentioned, and which source URLs are cited. Note factual errors about your company and the buyer constraint the answer addressed. Open the cited pages. A link beside your name does not prove that the page supports the claim.
Repeat the baseline quarterly using the same core questions and conditions. Add emerging questions in a separate group to preserve comparability. If capacity allows, run the panel twice per audit to expose variation. Ask: where are we absent, where are we described incorrectly, and which sources recur? Those answers should determine the work backlog.
Make important content readable without JavaScript
In a December 2024 study, the OpenAI, Anthropic, Meta and Perplexity crawlers tested by Vercel and MERJ did not execute JavaScript. Treat that as a dated finding about those crawlers, not a permanent rule for every AI engine. It is still a sound reason to make important content available in the initial HTML response.
Ask your web lead to fetch a service page and an article without running JavaScript. Can they read the actual explanation, author information and supporting evidence, or only a shell waiting for scripts? Use server-rendered HTML or static generation for essential content. You can retain interactive tools without hiding the text a buyer needs to understand your company.
Review robots.txt and firewall rules together. Confirm that the crawlers you intend to admit can reach your public pages without login barriers or bot challenges. Have legal and the web owner agree on access policy; do not remove every restriction to chase visibility. Check for accidental noindex settings and broken links. Allowing access creates eligibility, not a promise of citation.
Add accurate structured data describing the organization and articles where appropriate. It should match visible content. Then compare entity facts across your website, LinkedIn and relevant directories: company name, category, headquarters and service descriptions. Assign an owner to correct conflicting profiles. A firm described as an agency in one place and a software vendor in another is handing the reader an avoidable ambiguity.
Write passages a buyer can quote accurately
A useful answer should make sense outside the page that contains it. Put a direct response beneath a heading that matches the buyer's question. State the conditions under which the advice applies. If a reader needs six paragraphs of background to understand the claim, rewrite the passage before worrying about a new optimization tool.
Say a software vendor wants to answer 'How long does implementation take?' A generic promise about seamless deployment avoids the decision. A better passage explains what determines the schedule, who must supply the data and what can delay approval. Publish a defensible range only if your records support it. Explain exclusions so an engine cannot easily turn a conditional estimate into a universal promise.
Original data gives buyers something worth citing. Use permissioned customer evidence or your own documented analysis, with a clear method and limitations. Explain the sample behind a benchmark. Put the findings in readable HTML rather than leaving all the evidence inside a gated PDF. Publishing a number without its basis weakens the case when procurement asks where it came from.
Name the author and explain their relevant experience. Show publication and substantive update dates. Keep sources close to the claims they support. These are editorial standards you can control, not guaranteed ranking factors. Resist producing hundreds of lightly varied answers: give an expert responsibility for a smaller set of pages that address real evaluation questions.
Earn presence in the sources your audit finds
Your own site is only part of the evidence available to an engine. Use the cited-source inventory to identify where your category is being explained. Depending on the question, that might be an industry publication, a comparison directory or a practitioner discussion. Do not assume the same source matters across every engine or buyer problem.
For each recurring source, ask whether your company belongs there and what legitimate contribution it can make. Correct an outdated directory profile. Give an editor useful original research. Help a customer publish an honest implementation account when they are willing. Participate in practitioner discussions with a disclosed affiliation and an answer that helps the reader. Buying fabricated reviews creates evidence you cannot defend.
Assign this work to the people who already own those relationships. Public relations can pursue editorial coverage; customer marketing can support permissioned customer evidence. Have the organic search owner maintain the shared question and source inventory. I have seen marketing teams respond to a new channel by building another reporting silo. Keep search and answer-engine work under one plan, while retaining separate measures where they answer different questions.
Measure share of answers as a sample
Define 'share of answers' before reporting it. A practical definition is the percentage of eligible sampled answers that recommend your brand, counting each answer once. Track incidental mentions separately. Also track citations to your domain and material factual errors. A recommendation, a citation and a referral visit are different events. Do not collapse them into one visibility score.
Here is a hypothetical example. An audit produces 100 eligible answers to unbranded shortlist questions. Your company is recommended in 12, so its sampled recommendation presence is 12%. If a competitor appears in the same answer, both brands count. The brand percentages therefore need not total 100%. This is presence within a defined panel, not market share.
Report the counts beside the percentages, split by engine and question group. State which responses qualified and separately count queries that produced no answer. Retain refusals and failures in the audit log. Never redefine the denominator after seeing the result. A change concentrated in one engine or one question deserves a different discussion from improvement across the panel.
Answer engines are volatile. Wording, context and product changes can alter an answer even when your content has not changed. Show the range across repeated runs rather than a precise-looking decimal. Treat quarter-to-quarter movement as directional evidence, then inspect the underlying answers. Pair it with identifiable referral traffic and buyer interviews, but do not assign pipeline to an AI mention without evidence of that connection.
Give the first 90 days an owner
On Monday, ask sales for the questions buyers actually ask before agreeing to a meeting. Have your organic search owner build the panel and schedule a baseline review. Request the no-JavaScript page test from the web team. No new platform purchase is required to start.
Use the baseline to choose a few evaluation pages to improve and the recurring third-party sources worth approaching. Give each task an owner and an acceptance condition. 'Improve AI visibility' cannot be accepted as complete. 'Correct the implementation claim on these pages and verify it in readable HTML' can.
At Nova Nexus, we use Insight. People. Results. Read the evidence before changing the budget, establish who can fix the gaps, then report the work and observed outcomes. The Nexus Operating Review can help when ownership is unclear, but a CMO can start this program internally. Bring the next quarterly audit to the same review as organic search, with the saved answers available whenever someone challenges the numbers.
See how your operation compares.
Score your marketing operating system in five minutes, or talk it through with our founder.
Get new Insights by email.
New papers and points of view for CMOs, about once a month. No other mail. Unsubscribe any time.

