A sales rep gets an email: "We've compared three vendors and narrowed it down to you and one other. Two questions about implementation." It looks like a good lead. The problem is that the same comparison happened at dozens of companies that never wrote in, because your name wasn't on the list an AI chat gave them.
That is how artificial intelligence changes sales in practice. The headline isn't that reps draft emails with ChatGPT. It's that buyers now run part of the conversation a rep used to lead, on their own, with a language model. They set the criteria, compare offers and drop vendors before anyone at your company knows they were looking.
Buyers ask AI before they ask a rep
Gartner surveyed 645 B2B buyers in August and September 2025. Forty-five percent had used generative AI during a purchase, mainly to research vendors and products. On average they consulted about seven information sources per purchase (Gartner, 2026).
Retail shows the same shift. In an Adobe survey of more than 5,000 US consumers, 39% had already used generative AI for online shopping, most often for product research (55%) and recommendations (47%) (Adobe, 2025). Adobe sells analytics software, so read this as a directional signal rather than a neutral benchmark.
A working paper by OpenAI researchers and a Harvard economist found that practical guidance and seeking information are among the most common uses of ChatGPT, and that much of its value comes from decision support (NBER, 2025). Buying is a decision.
What this does to your pipeline
Buyers don't type a keyword into a chat. They write a sentence with constraints: "help desk software for a 15-person team, Salesforce integration, live within a month, under $500 a month." The answer is usually a short list with a reason next to each name. Follow-ups come next: "which is cheapest at 20 seats?", "what are the downsides of X?"
Three things follow for sales.
The shortlist forms without you. If your company isn't in the answer, you don't get the demo. You don't see the loss either, because it leaves no trace in your CRM.
Buyers arrive with a picture of your offer already formed. It may be accurate or not. If the model quoted an old price or said you lack an integration you shipped a year ago, the rep's first job is correcting the record. We cover this in AI got facts about your company wrong.
The sales call becomes a validation step. In Gartner's survey, 69% of buyers preferred to check AI-generated information with a rep, and trust was split almost evenly: 51% thought AI was more likely to mislead them, 49% picked the rep. Sellers aren't going away. Their job moves from presenting the basics to confirming facts and handling doubts.
Why AI recommends one company and skips another
An artificial intelligence chat with web access, such as ChatGPT search, Gemini or Google AI Mode, usually retrieves a handful of pages related to the question and builds its answer from them. For its own generative features, Google describes expanding a question into related searches and drawing on pages from its index (Google Search Central). Other platforms work differently in the details, but the principle holds: a model can only recommend what it found and understood.
In practice, you usually lose for one of three reasons:
- your page doesn't answer the condition in the question: the buyer asks about Salesforce, your page says "integrates with leading CRMs",
- no proof: a competitor states a price threshold, implementation time and named customers in your sector, while you say "fast and flexible",
- the model can't fetch your page: for example,
robots.txtblocksOAI-SearchBot, which ChatGPT search relies on (OpenAI).
The way a company describes itself is no longer read only by people. A model reads it too, then answers the buyer. A vague line that a human fills in from context is a blank for the model, so it reaches for a source that says something specific, often a competitor's page or a third-party roundup.
What sales and marketing can do about it
Start with the questions you already hear. Reps know what prospects ask on a first call and what stalls a deal. Those questions, in the buyer's words, are the best prompts to test. Our guide to choosing buyer prompts for an AI audit explains how to pick them.
Then ask them in ChatGPT, Gemini and Claude, in a fresh chat with search turned on. Note the date, the brands named and the sources cited. Answers vary between runs, so one check proves nothing. If three phrasings of the question return the same list without you, you have a pattern.
Finally, add one question to lead qualification: "How did you hear about us, and who else were you considering?" Some answers will be "ChatGPT". That's the only reliable way to see this channel in your CRM, because traffic from AI answers often doesn't show up as a separate source in analytics.
Where TextRank helps
Comparing one AI answer with your own page by hand takes an hour or two: open every cited source, list what it says, line it up against your copy. TextRank does this for a chosen question and model.
AI visibility monitoring tools count how often a brand appears in answers, which is useful across hundreds of prompts. TextRank goes a step further on a single question: it checks the sources the model actually retrieved or cited and compares them with your page at the level of entities, claims and specific passages. You see which words, conditions and descriptions the winning source used, and which of them your page lacks.
The output is a plan: what to keep, what to rewrite, where to add an answer, what proof to get, and which question to set aside because your offer doesn't meet its condition. Every recommendation points to the prompt, the model's answer and the source, so the rep, the copywriter and the product owner are talking about the same passage. You can see a full example in from prompt to page change.
TextRank doesn't promise the model will start recommending you. No one can honestly promise that, because the platform makes the call. What you get is a stated reason for every change and a way to compare results before and after. If, for a given question, the model cites only reviews and forums, the audit will say so, and rewriting your page won't be the first step.
Start with one lost conversation
Take the question from the last prospect who said they "compared options in ChatGPT" and run it in TextRank. One lost vendor list will tell you more about your sales than a general report on artificial intelligence trends.