How to Audit Your B2B Content for AI Search

Buyers in your market are using AI to do the research that used to happen on Google. They ask it to explain the category, surface the relevant vendors, lay out what criteria matter. All of that happens before they visit a single vendor website, before they speak to sales, before any of it shows up in your pipeline. By the time someone books a demo with you, they’ve already formed a view of the category and where your brand sits within it.

We studied this across hundreds of buying journey prompts in 14 B2B categories and found that 84% of them produce no brand citation. We call this the dark AI phase, where your brand isn’t cited in those conversations, but it’s still being judged by them.

This changes what your content needs to do.

1. It needs to influence how AI describes your category, your brand, and the criteria buyers use to evaluate competitors.
2. It needs to add enough new value to let buyers form opinions about you via AI, so that when they eventually land on your site later in the journey, they’re already engaged.

Across our research, we’ve consistently found that most B2B content wasn’t built for either of those things.

It was built to get found. That worked when search was the primary discovery channel, but if your content has nothing new to offer LLMs, it gets consolidated across hundreds of similar pages. Factor in the volume of AI-generated content being churned out across every category, and the bar for what counts as valuable has moved considerably.

The goalposts have now shifted from “will we rank for this and get found” to “is our content credible to AI and influential to humans.” In this guide, we’ll walk you through how to answer this exact question by conducting a content audit across your library.

Why Traditional Content Audits Are Breaking Down

The traditional content strategy was built on a specific playbook. You find a high-intent query, you provide the best possible answer to it, and you optimise it for visibility so it can rank and get traffic.

The old way to do audits helped you verify that you had done those things. Is the content structured clearly? Does it match the search intent? Is it optimised for secondary keywords? Is the technical health of your site in top shape? Tools like Ahrefs and Screaming Frog turned that checklist into a score out of 100.

This whole system existed to get content found so people can visit your website and grow your traffic, and for a long time, that was exactly the right thing to optimise for.

But with AI search, this has shifted. It’s more binary than a ranking signal. If your content doesn’t add a perspective, a finding, or a data point that isn’t already in its training data, it doesn’t hurt your ranking. It just makes you completely irrelevant to it. Traditional audits don’t catch any of that. They score whether content is built to rank, and have no mechanism for whether your content has anything worth saying.

When Is a Content Audit Worth Doing?

The teams that come to us asking about content audits usually want to know how often to run them. What we find is that the better question to start with is whether there’s a gap between who you are now and what your content says you are. When there is, an audit is worth doing. When it’s growing, it’s urgent.

And in our experience, that gap is almost always bigger than teams expect. Every piece of content you’ve ever published is still out there, and AI reads all of it simultaneously. It has some awareness of recency, especially when you signal it clearly, but newer content can still be drowned out by the volume of what came before. It synthesises across your entire library and builds a composite picture of who you are.

When your newest content reflects your current positioning and older content reflects something different, AI blends both. A buyer asking about you doesn’t get your latest thinking. They get a version of you assembled from everything you’ve ever published.

Running on a fixed review cadence means that gap goes unchecked for the full length of each cycle. The more you publish in the meantime without reviewing what’s underneath, the more these contradictions compound into what we call content debt. A page with no traffic and no recent updates isn’t invisible to AI. It’s still part of your footprint, still contributing to the picture AI builds of who you are.

There’s no specific checklist, but here are a few situations when an audit is worth prioritising.

• You’ve launched a new feature
• You’ve updated your core value prop
• A reposition up or downmarket
• A rebrand or name change
• Six or more months of publishing at volume with AI

Pre-Audit Preparation

Audits that produce unusable insights almost always fail for the same reason. The team jumped straight into evaluating content without first deciding what they were measuring, or how much of the library was in scope.

To ensure your audit generates a clear, valuable output at the end, there are a few things to sort out first.

Align with leadership first

If you plan to schedule a meeting with the CEO to walk them through your content audit proposal, that meeting isn’t happening. And if it does, it won’t be useful because stakeholders at that level don’t engage with content at the granularity an audit requires, and asking them to is the wrong entry point.

So first, go into the audit with genuine clarity on what the business needs content to do right now.

Is content primarily a lead generation channel? A pipeline influence tool? An AI visibility exercise? The answers to those questions determine the lens your audit uses. An audit run without that clarity will produce findings that don’t connect to anything leadership has a reason to act on.

Then, make sure the people who’ll need to approve changes know why the audit is happening before you go dark for two weeks.

The last thing you want is to surface from a detailed audit to find people asking why content output has dropped. The conversation doesn’t need to be long. Cover what content needs to do, why some of that isn’t happening, and what you’re going to look at and for how long.

Decide your audit scope

SEO and AI search have different requirements, and the first decision is which one is driving this audit. A piece can look like a problem through an AI lens but be pulling significant SEO traffic on a high-intent query. Audit from one lens only and you risk breaking something that’s working. Know your goal before you start.

Start with a bird’s-eye view of what you’re working with. You can use tools like Demand Genius or Screaming Frog to crawl your site, pull every indexed URL, and add columns for page type, topic, and funnel stage where you know them. If your priority is technical SEO and you’re comfortable in Google Sheets, Screaming Frog is probably best. If your priority is AI search and the ability to run flexible, judgement-based analysis, Demand Genius is the better fit.

From there, filter to what’s actually in scope. If your content library runs to hundreds or thousands of pages, triage it down to a working shortlist of 40 or 50 pieces before going deep. This is where the tool you choose matters. If scope is determined by the pieces with the biggest technical red flags, Screaming Frog is perfect. If it’s determined by more qualitative analysis (citability, extractability, differentiation, currency, quality), Demand Genius lets you deploy AI agents to assess all of those in minutes and flag the content that needs your attention.

There’s a lot of excellent content out there on how to conduct a technical SEO audit (we recommend X or Y). The remainder of this guide focuses on conducting a qualitative audit with AI search as the primary goal, while still making sure you don’t create SEO risk in the process.

Assess your current AI presence

Before pulling your data, it’s worth taking a look at how AI currently perceives and describes your brand.

If you don’t have access to any major AI visibility tools, there is a simple manual way to do this.

Pick ten queries that someone in your ICP would run during research, like TOFU queries when they’re looking for information, MOFU queries where they are trying to do a job, or BOFU queries where they’re looking for alternatives to your competitors.

Look at different aspects such as whether your brand was mentioned, how you are being mentioned, how you are being positioned against those competitors, or any hallucinations coming through.

This can give you a reasonable amount of data to gauge a baseline.

If you have access to AI visibility tools from the previous section that we mentioned, then it is a better way to understand this, because they give you deeper stats around metrics. For example, Demand Genius can tell you whether what AI says about your brand is accurate and whether it reflects your current positioning.

Even with that data though, a human still needs to make the call on which gaps are worth prioritising and in what order.

Pull your data sources

Data for a content audit sits across multiple tools and is rarely in one place. Before you evaluate anything, pull it together.

Start by getting every published URL into a spreadsheet. Screaming Frog, Sitebulb, or Demand Genius will do this in minutes. Add columns for page type, topic, and funnel stage where you know them.

From our experience, these are the most common data sources B2B teams pull from:

Before you start scoring, talk to sales and RevOps too.

A piece that barely registers in these tools might be getting forwarded in every competitive evaluation. Find out what they lean on, what they’d push back on changing, and what they wish existed. Content feeds everything in your GTM, and that conversation will tell you things the data above won’t.

Define your scoring criteria

Finally, you need to decide what good looks like.

Set up criteria based on the goal you had at the start of the preparation. Decide what sort of metrics would determine whether your content is performing well, poorly, or very poorly.

Certain factors will carry more weight than others, and that’s what you’re trying to understand. When you define these dimensions and their relative importance, you can prioritise the audit correctly and see where the real gaps are.

For each dimension, think about how you want to score it. You can score on a numerical basis, or you can use different levels, like high, medium, and low. As long as you have that set out, everything downstream is going to be straightforward to audit.

For example, we recently conducted an audit of a large B2B content library. At the beginning, we noticed that they had a lot of content that was being pushed out using AI, which is always very generic content that does not add any differentiated value.

So one of the important factors we decided to score on was information gain, which explains whether their content is actually adding anything new for LLMs or not. This was important to analyse because we’ve seen from our AEO research that if there is no new information being added, then there is no reason for AI to surface it.

If you want inspiration, you can take a look at this case study to get an example of how we set up scoring criteria for them.

How to Do a B2B Content Audit for AI Search

These five dimensions come from patterns we’ve observed consistently across our research into B2B content libraries. LLMs don’t evaluate individual pieces in isolation. They synthesise across your entire library simultaneously, building a composite picture of who you are. It’s rarely just one dimension causing the problem — it’s usually a combination of two or more.

In the sections below, we go through each of the five dimensions in turn, with practical guidance on how to audit for each one yourself.

Strategic Differentiation

Of the five dimensions, this is where we’ve observed most B2B content libraries have the biggest gap, and it’s also the one that hasn’t had any real pressure to address seriously until now. As long as your technical signals were right and you were publishing content optimised for ranking, you could get away with it and still drive traffic.

Now, AI is able to answer a user’s query by synthesising information across ten different articles. If all of them are publishing the same ideas, they are all going to be consolidated and packed together for the reader. That way, the reader does not have to go through ten different articles.

We also performed an analysis of 90,000 HR tech content pieces to vet this, and what we found was that only 4.7% of all that content has actually shown a high level of differentiation. The remaining major chunk is just ideas that have been reiterated again and again, with nothing novel contributed to the category.

Within strategic differentiation, there are two different aspects to work through.

The first is information gain, which is a measure of how much new insight your content actually introduces. We find it useful to think about this in four levels, which we’ve covered in depth in this article.

When working through your library, ask this for each piece. What does this tell me that an LLM could not produce on its own? Using the levels we’ve established above, that can help you determine a score.

The second aspect is narrative ownership.

Imagine two content assets covering the same topic. One is framing it from a different perspective, while the other is using pretty much what most of the other competitors are doing. If you have strong narrative ownership, your explanation, your point of view, and your position on how things work are distinctively yours.

To assess this, pull up each piece and read it without looking at who published it. Could the byline belong to any competitor in your category without changing anything about the content? If yes, it is generic. If the framing and perspective are distinctively yours, it has ownership.

You can have Claude run analysis on each URL, or if you want to exponentially speed up this analysis without working through it piece by piece, you can use Demand Genius to set up an AI agent with each prompt and run it across every single content asset in your library automatically.

Content Quality

For your content to be worth drawing from, it needs to go further in at least one specific way. Either explaining concepts for buyers who are still forming their category knowledge, or covering a topic at enough depth that there’s something to learn from it. Unfortunately, from our research, we’ve seen that both are consistently weaker across B2B libraries than teams expect.

The three sub-factors to assess are definition clarity, content depth, and keyword optimisation.

• For definition clarity, read through a sample of your pages as if you’re encountering the category for the first time. Are key terms explained, or do they appear as assumed knowledge? A piece on demand generation that uses “dark funnel” or “pipeline velocity” without connecting either concept to anything is accessible only to buyers who already know the vocabulary. Those pieces work for an existing customer base, but not for buyers earlier in the journey, or the model serving them.
• To analyse content depth, work through the subheadings of each piece and ask whether any section actually teaches you something concrete. Not just what a concept is, but how it works or where it tends to fail. If every section is a brief overview that stops before things get specific, the piece is thin regardless of how well it reads.
• Keyword optimisation tends to reveal itself on a straight read. Forced repetition shows up immediately, and headings that restate the same query in slightly different forms for a crawler are a clear signal. Natural writing is what you’re looking for in this check, and worth protecting when you find it.

In practice, working through 40 to 50 pieces manually takes much longer than an hour. For most teams, a library of that size runs closer to a full week of focused work, and that’s before anything that needs a follow-up decision.

At that pace, you need AI to do this at any real scale. The question is how. You can put something together with ChatGPT or Claude, and it’ll work, but you’re building the evaluation framework from scratch, outputs are hard to reproduce consistently, and there’s no clean way to compare scores across pieces. Demand Genius is built for this specifically. It comes with a pre-optimised template library for content quality assessment, runs the three sub-factors across your full library automatically, and gives you scored results in one view you can act on.

You can see where the worst-performing content is concentrated and prioritise from there. There’s a walkthrough below on how to set it up.

AI Extractability

Well-written content can still be hard for an LLM to use. Extractability measures how easily a model can pull accurate information from your content when building a response.

The fastest test is to read only the headings and the first sentence of each section in a piece. If the main argument holds together through that reading alone, the structure is working. When you have to read into the middle of a paragraph to understand what a section is actually about, key information is buried.

The second thing to check is directness. Specifically, whether the most important claim in each section is stated explicitly or implied through the surrounding argument. Find the key claim in a section. Is it the first sentence, or is it something you have to infer after reading the whole thing? High-extractability content leads with the insight. The supporting context and evidence follow.

It’s worth checking both separately. A piece can be well-structured with clear headings and logical flow, and still be indirect in how it states its conclusions.

Citation Likelihood

Our research has found that only 16% of AI responses in a B2B buying journey produce any brand citation, which is concentrated on high-intent queries like evaluations, comparisons, searches where a buyer is already testing a shortlist. For most of the journey, LLMs are shaping buyer understanding without citing anyone.

Regardless, citation likelihood matters as an audit dimension for a specific reason. The content that tends to get cited and the content that does the most work in the dark AI phase share the same underlying property: both make clear, well-evidenced claims. Running this check tells you how much of your library has enough substance to pull weight in either place.

To do this, start by analysing brand assertion, which is the strength and specificity of claims. Read through the content pieces looking specifically at how statements are made. Assertive content makes specific, verifiable claims. A number, or a named outcome that can be checked. Hedged content qualifies its way to safety using phrases like “aims to,” “can help,” and “is designed to.” Given the surge in AI-generated content, we’ve noticed many B2B content libraries accumulate this over time and experience terrible results.

Then look for authority signals. What you’re looking to understand is whether the content backs up what it claims. Supporting evidence such as original data or case study examples raises the weight of a piece as a source. Pieces without any of it are making claims the model has no stronger reason to draw from than any other piece covering the same topic.

Accuracy and Currency

Outdated content in your library is the most direct source of AI responses that misrepresent your brand. In search, old content that stops getting traffic mostly decays by dropping in rankings. LLMs don’t work this way.

A piece published three years ago, reflecting positioning you’ve since moved on from, is still part of what they draw on when describing your brand. It contributes regardless of traffic and regardless of when it was last updated.

Here’s how to assess for this dimension.

1. Look at how current and time-relevant the information is, relative to the current year. The goal is to categorise each piece as one of three things: Current (data and references are up to date), Evergreen (foundational content that stays accurate regardless of when it was published), or Outdated (contains old statistics, obsolete references, or examples tied to conditions that no longer apply). Most libraries contain all three, and separating them tells you where refresh effort actually needs to go.
2. Determine whether the content explicitly references any of your products or brand by name. Not every piece will, but where it does, those references become the source material LLMs draw on when describing what you offer.
3. Validate whether those references match your current brand positioning. A piece can be current in its data and still be describing a version of your product that no longer exists. Cross-referencing mentions against your current positioning is how you find the pieces that are actively working against you.

Auditing across these three checks can reveal which pieces need a refresh and which ones are genuinely evergreen.

What we’ve seen is that this is difficult to stay on top of because brands keep changing. Positioning shifts, products evolve, and it’s hard to catch every piece of content that’s been left behind. In Demand Genius, you can set up an AI agent that is specifically looking for accuracy and currency automatically. That way, every time your brand goes through any positioning changes, the agent keeps auditing your full library against it and flags anything that looks outdated or misaligned right away.

How to Analyze Your Findings

Once you have your audit findings in hand, you know where things are falling apart and how your content is showing up overall. The challenge is that most audits never get acted on. They produce findings, and then those findings sit. So before anything else, you need a clear framework for how to move through what you’ve found.

Start by building a prioritised list based on the five dimensions you’ve just assessed. Combine all of that into a single overall score per piece. Set a cut-off. Below it, pieces get flagged for action. Everything above stays in maintenance mode.

Before deciding what to do with the flagged pieces, overlay them with traffic data. Some of them are going to score poorly on the audit dimensions but still be pulling significant traffic. That’s not a simple go-and-update situation. You don’t want to put that traffic at risk, but those pages still need a structural lift and the fixes they need for AI search. Find the balance with those pieces and be careful.

For each flagged piece, assign one of four actions.

Lily Ray recently posted about the Mount AI trajectory the industry has been following since AI came through in 2024, and we saw the same pattern in our own HR tech study. Content velocity grew roughly 8x in a year as brands scaled up AI-assisted production. And yet overall performance regressed. The pattern was consistent - a brief visibility spike, then everything crashed back to where they started.

A big part of what drove this was producing content without differentiating it, which ties directly back to one of the five dimensions we’ve covered above. Teams kept pushing out new content without maintaining what was already there.

So when you’re building your roadmap, keep that in mind. The pieces you’re working on should focus on genuine differentiation, not just churning content for the sake of it. Build out a 90-day roadmap that works through priorities from highest impact to lowest.

And follow a rough 50/50 split as you do. Put 50% of your effort toward new content and 50% toward updating and maintaining what’s already out there. Every net new piece you plan on creating adds more ongoing work for yourself.

Then, set clear owners with target dates and make sure things don’t get lost. Use a project management system, whether that’s Asana or whatever your team is already in, to hold accountability and keep work moving through the pipeline.

As work ships, track the results. Go back to the success metrics you defined in pre-audit preparation and start monitoring what’s changing. What you’re tracking is whether AI is describing your brand more accurately, citing you more often on the queries that matter, and whether that’s translating into pipeline.

Switch to Continuous Auditing with Demand Genius

If there’s one thing that should be clear after reading this, it’s that a content audit is no longer a project you do on a target timeline or by a certain due date. For B2B brands that want to dominate AI search, this needs to happen on a much more frequent cadence. At this point, it’s almost mandatory, because AI platform behaviour is constantly changing.

A quarterly or annual audit isn’t nimble enough to catch the issues that matter like content decay or competitor gaps. You can’t just be reactive and deal with it when something breaks. You have to be proactive, and that’s only possible when you’re consistently on top of what’s changing.

The challenge is that doing this continuously is resource intensive, and most teams simply don’t have the bandwidth. That’s what we set out to build when we started Demand Genius.

Demand Genius has been built exclusively for B2B teams. If you have complex buying behaviour, multiple stakeholders, and a content library that’s constantly evolving, you can use Demand Genius to set up automated continuous monitoring that is tailored to your use case and your product. It flags your content on your choice of dimensions such as when it starts degrading in AI visibility or when you want to keep track of whether your content stays differentiated over time.

If you’d like to get this set up for your brand, [book a demo here].

FAQs

How long should a content audit take?

If you’re doing it manually, 50-100 URLs take about 6-10 hours over 2-3 days, 100-300 URLs take around 12-20 hours, and enterprise sites (300+ URLs) can take 2-4 weeks for a comprehensive audit. But if you use Demand Genius, you can set up smart AI agents that are dedicated to analysing content for each factor you want to assess. This method, in comparison, takes a couple of minutes to an hour or so, depending on site size. Plus it’s continuously auditing. You just set it up once.

Who should own the audit?

Ideally a content strategist or SEO lead. But it’s also important to get supporting inputs from marketing, sales, RevOps, and product as necessary.

How often should B2B teams audit their content?

The most common cadence is either quarterly or annually. But AI search engines update much faster than Google’s algorithm, so this older cadence is obsolete. Now you need either a monthly audit as a baseline, or better yet, constant automated checks that analyse your footprint on the fly.

Can AI tools run a content audit?

As long as you set it up properly, AI is pretty good at doing hard analysis that is often very time-intensive, like quality scoring at scale, tracking your overall footprint, finding gaps, etc. But you still need humans for strategic judgement, looking for brand specifics like brand positioning or pipeline fit.

What if our content is fine on SEO but underperforms in AI search?

There’s a good chance your content is still doing the old jobs like keyword optimisation and internal links, which are still important, but it can do better in terms of structure for LLMs to pick it up, like FAQ schema. If you’re using AI for content production, it might have AI writing patterns like hedged language, generic ideas, etc., or it may not be covering the deeper aspects that LLMs draw on when generating responses for a particular topic. What you want to do is ensure it follows the new best practices you need to succeed at performing in AI search.

What if we are missing CRM attribution?

That’s not an ideal scenario, especially if you have complex sales cycles, but use whatever you can get your hands on, like GA4 or assisted conversions if you have them. Make rough estimates and work forward from there. You can also involve sales to analyse qualitative signals that can reveal how prospects might be discovering your content via AI search and mentioning it to reps, which could be influencing deals.