How do I turn AI visibility data into a content plan for demand generation?

To turn AI visibility data into a content plan for demand generation, start with the buyer questions that sit closest to your pipeline, read which evidence gap each lost answer shows, and write one brief per gap that names the facts and sources the page needs. Then order the briefs by effort and by buyer stage, fit them to your campaign calendar, and measure the same questions again once the work is live. The plan is only as good as the measurement under it, so the questions and the counting stay fixed from one round to the next.

Most teams that measure their AI visibility end up with a spreadsheet of answers and no clear next step. This playbook is the bridge from that spreadsheet to a plan a demand generation team can run. It does not repeat two pieces it depends on. Building a buyer-question set covers how to collect, weight and freeze the questions, and Evidence gaps describes the twelve gap classes and the first fix for each. Here the job is different: deciding which gaps matter for pipeline, turning each into a brief, and putting the briefs in an order your team can deliver.

Where this sits in the method

We use the same five steps on every piece of AI visibility work: measure the questions, read the evidence, find the gaps, act, and measure again. How a B2B tech company gets recommended more often by ChatGPT and Claude walks through all five for two assistants. This playbook zooms in on the turn from reading to acting, which is where demand generation comes in.

The measurement is assumed to exist already: a frozen set of buyer questions, each asked of each assistant more than once, with every answer recorded by date, assistant, the brands named, the role each played and the sources cited. If you have only a handful of answers, read them as a direction. Reading an AI visibility figure honestly explains how much a small base can bear.

One fact about the data itself shapes everything that follows. OpenAI's help article on searching the web with ChatGPT says answers that use web search may include citations and a list of sources, and also that search results and citations can be incomplete, out of date or wrong. Anthropic's help article on web search in Claude says answers that search include citations so a reader can check them. So the citations in your data are a real view of what an answer rested on, and an imperfect one. The plan below treats them as evidence to read, not as a complete record.

The worked example

The example runs through every step. It is fictional: a UK HR software company that sells to employers of 200 to 2,000 staff, whose strengths are payroll integration and absence management. It tracks two rivals, Rival A and Rival B. Its demand generation team runs two campaigns a year, one ahead of the payroll year end and one on absence management, and its sales team creates most opportunities after a buyer has built a shortlist of three or four vendors.

It has measured 15 frozen questions on two assistants, twice each, so it holds 60 recorded answers.

Step 1: Choose the questions that matter for pipeline

Not every lost answer deserves a page. Start by ranking the questions you measured by how close each one sits to a deal. Three readings help.

• Where deals start. Ask sales which questions prospects raise in first calls and which constraints appear in won and lost deals. A question that mirrors a common first-call question is close to pipeline.
• Which campaign it feeds. Mark each question with the campaign or programme it belongs to. A question no campaign touches can still matter, but it will not have a team behind it.
• The stage it sits at. Questions that compare named options or check one vendor sit nearer a shortlist than questions that only describe the problem. Keep both, but weight them differently.

The weights you agreed when the question set was frozen are the starting point. Do not re-weight after reading the answers, or the plan will drift towards the questions where you already look good.

• Inputs: the frozen question set with its weights, the answer record, sales input on first calls and deal reasons, and the campaign calendar.
• Outputs: a shortlist of questions ranked by pipeline value, each tagged with its buyer stage, the role it speaks to and the campaign it feeds.
• What this cannot tell you: how much pipeline any AI answer has influenced so far. The ranking says where a lost answer would hurt most, not what it has already cost.

In the example. The team shortlists 6 of its 15 questions. The payroll integration question and the "alternatives to Rival A" question rank highest, because both mirror what prospects ask in first calls and both feed the payroll year-end campaign. A question about how to cut sickness absence ranks lower: it sits at the problem stage, and buyers asking it are rarely ready for a shortlist, though it feeds the absence campaign.

Step 2: Read which gap each lost answer shows

For each shortlisted question, take the answers where you were absent or named but not put forward, and write one line per lost answer: the question, the assistant, the shape of the loss, the gap class or classes, the rival that was put forward and the source it rested on. Use the classes in Evidence gaps rather than inventing new ones, so the plan speaks the same language every round.

Read the answer before you classify it. Several gaps often apply to one answer, and some answers fit no class; record those as unclassified rather than forcing them into the nearest one.

• Inputs: the lost answers on the shortlisted questions, with their citations.
• Outputs: a gap register: one line per lost answer, with its gap classes and the winning evidence.
• What this cannot tell you: whether the gap you recorded is what the assistant actually needed. It is a careful reading of what was missing, not a view inside the assistant.

In the example. On the payroll question, the register shows a coverage gap on both assistants (no page of the company's on the question), a specificity gap where Rival A was put forward (its page names the UK payroll systems it connects to), and a corroboration gap on Claude (a UK directory listing backs Rival B, and the company's listing is out of date). On the "alternatives to Rival A" question, the company is named in 2 of 4 answers but never put forward, and the answers that name it cite only its own homepage.

Step 3: Write a brief per gap, with the evidence it needs

This is where AI visibility data becomes content. Each gap, or each group of gaps that one piece of work would close, gets a short brief. A brief that a writer, a web editor and a demand generation manager can all act on has nine fields.

1. The question, in the buyer's exact words, and the assistants where it was lost.
2. The gap, by class, with the line from the register that shows it.
3. What won, meaning the source the winning answer cited and the claim it supported.
4. The facts the page must state, specific and checkable: systems, standards, sizes, regions.
5. The proof, meaning the citable page or document that backs the main claim.
6. The independent source, meaning the listing, directory profile or comparison to update or request.
7. The format, whether a new page, a rewrite, a listing update or a correction, and where the text must sit so a crawler can read it.
8. The campaign it joins, so demand generation can use the same page in an email, an ad or a sales sequence.
9. The owner and the re-measure date.

The brief never asks for the brand name to be repeated more often or for wording aimed at a model. It asks for the evidence a careful buyer would want, stated plainly, because that is the evidence an answer can use.

• Inputs: the gap register, the winning sources, and your own product facts.
• Outputs: one brief per gap or group of gaps, each with an owner and a date.
• What this cannot tell you: which sentence an assistant will quote, or whether it will quote the page at all.

In the example. The payroll brief asks for one page titled with the payroll question. It must name the six UK payroll systems the product connects to, say which sync in both directions, and link to the integration documentation as proof. It asks for the UK directory listing to be updated with the same six names, and for a short correction to go to the author of the comparison article. It joins the payroll year-end campaign, is owned by the content lead, and carries a re-measure date eight weeks after publication.

Step 4: Sequence by effort and buyer stage

With the briefs written, put them in an order your team can deliver. Sort them on two readings.

• Effort. Access fixes and corrections are small. A rewrite of an existing page is medium. A new page with its proof is larger. An independent source you have to earn is the slowest, because someone else decides when it changes.
• Stage and weight. Briefs for questions near a shortlist, with high weights, go before briefs for problem-stage questions.

Then fit the order to the campaign calendar. A page that answers a question a campaign is about to raise should be live before the campaign starts, so the same evidence serves the buyer who reads the campaign and the buyer who asks an assistant. Do access fixes first, whatever their stage, because a page no crawler can reach helps no one.

Keep the plan to what the team can ship in a quarter. A few briefs done well beat a long list half done.

• Inputs: the briefs, the team's capacity, and the campaign calendar.
• Outputs: a dated content plan for the quarter, brief by brief, with owners, launch dates, linked campaigns and re-measure dates.
• What this cannot tell you: that this order is the best one. It is a reasoned order, and the re-measure is what tests it.

In the example. The plan for the quarter has five items. First, a small fix: the integration documentation is behind a login, so it is opened to the public. Second, the payroll page, timed to go live three weeks before the payroll year-end campaign. Third, the directory listing update and the correction, sent the same week. Fourth, a comparison page for buyers weighing alternatives, with the facts a buyer would check. The absence management page waits for the next quarter, because its question sits at the problem stage and the team has capacity for four items.

Step 5: Measure the same questions again

When the work has been live long enough to be found, ask the same frozen questions again, of the same assistants, the same number of times, and record the answers the same way. Compare question by question and assistant by assistant against the baseline. Report the questions that did not move as plainly as those that did, and turn what you learn into the next quarter's briefs.

Read pipeline signals beside the measurement, not as proof of it. A form field asking how a buyer found you, with an AI assistant as one option, tells you something. So do sales notes where a prospect says an assistant suggested you. Some visits from assistants may also show up as referrals in your analytics. None of these is a full count: a buyer who reads an answer and later searches for your name, or types your address, arrives with no sign of the assistant at all. Treat the signals as a lower bound on influence, and never add them up into a figure the data cannot carry.

• Inputs: the frozen question set, the baseline, the change log, and whatever pipeline signals you already collect.
• Outputs: a before and after per question and per assistant, a short note of pipeline signals that mention an assistant, and the next quarter's gap register.
• What this cannot tell you: how much revenue AI answers produced, or that your work caused a change. A provider can change its assistant at any time, and rivals publish too.

In the example. Eight weeks after the payroll page went live, the re-measure shows the page cited in 1 of 2 ChatGPT answers to the payroll question, and the company put forward as an alternative in 1 of 2 Claude answers. The "alternatives to Rival A" question has not moved. Three inbound forms in the quarter name an AI assistant as how the buyer found the company, which the team reports as three forms, not as a share of pipeline. The comparison page and the absence page head the next quarter's plan.

Where the plan meets demand generation

The steps above produce a content plan. What makes it a demand generation plan is that every brief is tied to the pipeline from the start.

• Shortlist questions feed sales. Pages that answer evaluation and decision questions are the same pages a salesperson can send after a first call. Give sales the list as each one ships.
• Problem questions feed programmes. Pages that answer problem-stage questions belong in nurture and content programmes, where they earn attention before a buyer has a shortlist.
• Campaign timing serves both audiences. When a page goes live before the campaign that raises its question, the buyer who clicks the campaign and the buyer who asks an assistant meet the same evidence.
• One owner reads both sets of numbers. The person who owns the plan should see the AI visibility record and the pipeline signals together, so neither is read alone.

For demand generation and RevOps teams, our page for demand generation and RevOps covers how AI answers shape shortlists and how to read AI-referred inbound without over-claiming it.

What this cannot promise

• No pipeline figure. The plan ranks gaps by where they would hurt most, and the signals in step 5 show some of the influence. Nothing here measures how much pipeline AI answers create, and no honest plan should claim it does.
• Answers vary from run to run. A change on a few answers can be ordinary variation, which is why the method asks each question more than once and compares like with like.
• No one outside the providers controls what an assistant recommends. A content plan improves the evidence an answer can draw on. It does not decide the answer.
• Timing is unknown. We know of no reliable, published measure of how long a new page takes to appear in an assistant's answers, so re-measure dates are set to give the work a fair chance, not because a result is due on that date.

If you want this run for you

The method can be run in-house with a spreadsheet and a disciplined team. If you would rather have the measurement and the gap reading done for you, the AI visibility audit maps the evidence gaps on your buyer questions and gives you a prioritised plan, and AI visibility measurement and marketing repeats the measurement each month and does the work the briefs call for.