Sales and marketing software is the hardest category in B2B for getting named by an AI engine. Not because the products are worse or the marketing is weaker. Because the category has a few enormous incumbents, forty companies describing themselves in the same dozen phrases, and comparison content owned almost entirely by people who are not vendors.
That combination produces a specific failure. Ask an engine for the best sales engagement platform and it will name the same four companies it named last month, sourced from review aggregators and listicles rather than from anybody’s product pages. Your content quality has almost nothing to do with it.
This is the playbook we run for companies in this category. It is ordered, because the order matters more than any individual tactic.
Why this category behaves differently
Three structural facts, and none of them are about your website.
The incumbents are over-represented in training data. A company that has existed for twenty years, been written about constantly and filed public documents is dense in any model’s parameters. A company founded in 2022 is not. That asymmetry is not addressable by publishing, and pretending otherwise is how budgets get wasted.
The comparison layer is not yours. In most categories, vendors publish the comparisons. In this one, review platforms, agencies and affiliate sites do, at enormous volume. When an engine retrieves comparison content about sales tools, it retrieves theirs. Your own comparison page is one voice among hundreds and reads as self-interested, which is the correct reading.
Everyone uses the same words. Pipeline. Engagement. Revenue intelligence. Go-to-market. These phrases carry almost no discriminating information, which matters because dense retrieval works on meaning. Forty companies whose descriptions embed to nearly the same point in vector space are, to a retriever, interchangeable.
The consequence worth internalising. In this category you are not competing to be described well. You are competing to be distinguishable. Those are different problems and the second one is mostly solved by narrowing, not by writing more.
The four positions actually available to you
There is a useful way to think about where you can realistically be named, and it is not a funnel. It is a grid of two questions. Does the buyer name a category, and does the buyer name a constraint.
| No constraint named | Constraint named | |
|---|---|---|
| Broad category | Owned by incumbents. Not winnable in under two years. Do not budget for it | Winnable in one to two quarters. This is where the work goes |
| Narrow category | Winnable, low volume, high intent. Cheapest win available | Usually uncontested. Own it completely and quickly |
A constraint is anything that narrows the field. Company size, industry, a named integration, a compliance requirement, a workflow, a region, a team structure. The moment a buyer supplies one, the engine has something to discriminate on, and discrimination is what you need.
The practical move is to write for the constrained question rather than the category question. Not the best sales engagement platform, but sales engagement for a twelve person team selling into hospitals. That is a real question people ask and almost nobody has written a clean, retrievable answer to it.
The fastest way to find your constraints
They are already in your closed-won data. Look at what your last forty customers had in common that your lost deals did not. Company size band, one integration, one industry, one team shape. Two or three usually explain most of your wins.
Those become the qualified questions you write for. If you would rather see which of them engines already associate with you, a free visibility check tests the qualified variants alongside the broad ones and shows you the gap.
What buyers in this category actually do
Worth grounding this in how the buying works, because the sequencing follows from it.
Sales and marketing software has an unusually long, unusually self-directed evaluation. Buyers research extensively before contacting anyone, and by the time they do, the shortlist mostly exists. Salesforce’s recurring State of Sales research has tracked the same direction for years, which is buyers doing more of the work themselves and expecting information to be available without a conversation.
The mechanism underneath that is better explained by category research than by funnel diagrams. The Ehrenberg-Bass tradition is built on a finding that transfers almost perfectly to AI search. Buyers do not evaluate the whole market. They retrieve a small set of brands that come to mind for the situation they are in, and then choose within that set. Being in the retrieved set is most of the battle. Being best is a smaller part of it than anyone wants to believe.
An AI engine does the same thing, mechanically rather than psychologically. It retrieves a small candidate set and selects within it. Same structure, different substrate. Which is why the demand-side measure that correlates with market share in conventional marketing, tracked by the IPA and others as share of search, has a direct analogue here. How often you are retrieved matters more than how well you are described when you are.
The order of work, which is the whole playbook
Do these in sequence. Skipping ahead is the most common and most expensive error in this category, because step four is fashionable and steps one and two are not.
| Step | The work | Time to signal | Skip it and |
|---|---|---|---|
| 1. Access | Confirm AI crawlers reach you. Logs, not robots.txt | Days | Everything downstream is invisible and you will not know why |
| 2. Extractability | One passage per question, self-contained, no client-side rendering | Two to six weeks | You are retrieved for nothing regardless of authority |
| 3. Discrimination | Qualified pages for your two or three real constraints | Four to ten weeks | You embed to the same point as forty competitors |
| 4. Corroboration | Independent sources describing you consistently | Two to four quarters | You are quotable but never recommended |
| 5. Freshness | Pricing, integrations, leadership, positioning kept current | Ongoing | Engines describe a version of you that stopped existing |
Step one deserves a sentence of insistence. We find blocked or throttled AI crawlers on roughly a third of the sites we look at, and almost nobody chose it. It arrives with a bot management rule or a security plugin default. Checking takes ten minutes and the check is server logs, not the robots file, because the robots file tells you your intent rather than what happened. The stage model behind this is in indexed, retrieved, cited.
What to publish, specifically
Four page types carry most of the weight in this category. They are not exciting.
- A pricing page a machine can read. Pricing is the most-asked vendor question and the one most often answered from a stale third-party listing. If the number is not in your HTML as text, it does not exist. The full treatment is in why your pricing page is the page AI reads most.
- Integration pages, one per major integration. These are the highest-intent constrained questions in the category and they are usually handled with a logo grid. A logo is not a passage. One page that states what the integration does, what it syncs, what it does not, and what it costs will be retrieved far more often than the grid it replaces.
- An honest alternatives page. Not a comparison table where you win every row. A page that says plainly who you are not right for. Counterintuitively this gets retrieved and quoted more, because it is the only page in the category carrying disconfirming information.
- A glossary of your own terms. This category invents vocabulary constantly and buyers cannot tell a standard from a vendor coinage. Defining them, including marking which ones are yours, is cheap and gets quoted. The method is in the glossary page almost nobody builds.
Before you commission any of this, find out which step you are actually stuck on.
The five steps have completely different costs. A company stuck at step one needs a configuration change and a fortnight. A company stuck at step four needs two quarters and a budget. Buying the second when you need the first is the most expensive mistake available in this category, and it is extremely common.
Our free AI visibility check names which step is binding, with the crawler log read and the verbatim engine answers attached. You keep the output whether or not we ever speak again.
Measurement that survives contact with a board meeting
The temptation is a single visibility score. Resist it. In a category with this much incumbent gravity, a blended number will barely move for two quarters and you will lose the budget before the work matures.
Measure four things separately, because they move at different speeds and have different causes.
| Measure | What it tells you | Expect movement in |
|---|---|---|
| Named on qualified questions | Whether discrimination is working | One quarter |
| Named on broad category questions | Whether corroboration is working | Two to four quarters, or never |
| Accuracy of what is said | Whether freshness is maintained | Weeks |
| Source attributed to you | Whether extractability is working | Four to eight weeks |
Run the same fixed prompt set each month rather than inventing new questions, and keep the transcripts. The protocol is in the twenty prompt set, and the only modification for this category is that at least eight of your twenty should carry a constraint, because that is where your winnable ground is.
What we would not bother with
Chasing the category question. If three companies with a decade of press coverage own it, you are buying a lottery ticket with a content budget. Take the constrained questions instead and let the category question arrive later as a by-product of owning enough of them.
Publishing more comparison pages. The comparison layer belongs to third parties in this category. One honest alternatives page is worth six competitor-versus-competitor pages, all of which read as marketing to a reranker and to a human.
Paying for mentions. Sponsored listicles and paid placements in roundups do get crawled. They also get recognised as promotional, they cluster on domains whose whole business is that inventory, and the signal decays the moment the payment stops. Corroboration works because it is independent. Buying it removes the property that made it work.
A weekly publishing cadence with no constraint behind it. Volume is not the lever here. Forty undifferentiated posts embed to the same undifferentiated point as your homepage. Four pages that each own a constrained question will outperform them, and cost less.
A realistic twelve week plan
Weeks one and two, crawler access and rendering. Pull thirty days of server logs, confirm which AI crawlers actually reached you, fix whatever is blocking them, and check that your pricing and integration pages contain their key facts in the initial HTML response.
Weeks three to six, extractability. Take your ten most commercially important pages and rewrite so each one contains a self-contained passage answering each question you want it to win. No opening sentence that depends on the paragraph above it. This is editing, not commissioning.
Weeks seven to twelve, discrimination. Identify your two or three real constraints from closed-won data and build a proper page for each, plus the integration pages that support them. Done properly this is where you start appearing on qualified questions you previously lost.
Corroboration starts in parallel and finishes long after. It is the slowest and least controllable part, and it is what eventually moves the broad category question. Treat it as a two-year programme with quarterly checkpoints rather than something with a completion date. MIT Sloan Management Review has published a good deal on why organisational patience is the scarce input in this kind of work, and that is the honest framing to take into the budget conversation.
Related reading
- What AEO actually is. The general picture, if you want the mechanism before the vertical specifics.
- How ChatGPT picks which companies to name. Why incumbents dominate unqualified questions, in more detail.
- How Perplexity chooses its sources. The engine where a newer company can move fastest, and why.
- What AEO agencies charge. If you are working out whether to run this internally or buy it.
Start with the log check. It costs an afternoon and it reorders everything else.
In a category this crowded, the companies that move are almost never the ones that published most. They are the ones that found out they were blocked, fixed it, and then wrote four pages that said something specific enough to be worth retrieving.
If you want that first step done properly and evidenced, send us the domain for a visibility audit. We will tell you which of the five steps is binding and what it would take to clear it, including when the answer is that you do not need an agency.
Shaban runs answer engine optimisation for B2B SaaS companies at Uncited Brands. He works the unglamorous end of the problem, mostly crawler access, retrieval diagnostics and measurement baselines, and publishes the tests that do not go his way alongside the ones that do.