Sending isn't selling

"If those systems are fragmented, the agents will scale the fragmentation."

That's Dan Gottlieb, a VP Analyst at Gartner, in July. It's the most useful sentence anybody has written about AI in sales this year, and it went almost entirely unremarked because it isn't the kind of thing that gets shared.

The prediction attached to it is the bit that travelled. Gartner have AI agents outnumbering sellers by ten to one by 2028, and fewer than 40% of sellers saying those agents improved their productivity. That's off a survey of 210 chief sales officers and senior sales executives, run in January and February this year.

Read those two numbers together. Ten times as many agents. Under 40% of the people they're meant to be helping saying they helped. That is not a technology problem.

I Just Need Leads

I hear a version of this constantly and it always arrives dressed as progress.

We've got the agents running. Forty thousand emails a month. Twelve sequences live. Full coverage of the ICP.

Those aren't the metrics I want to hear. I can press send. Anybody can press send, and now the software can press send for you at a volume no human could match. So the real questions are the same ones they've always been. How many conversations did it start? How many deals closed? How much pipeline did it actually influence, as opposed to touch?

And here's what happens in practice, which is why Gottlieb's line matters so much. You point an agent at a CRM where close dates have moved four times, where there are three versions of the same account, where half the opportunities haven't had a real activity logged in a quarter. The agent doesn't fix any of that. It works faster than you inside it. So you get more activity, more emails, more logged touches, and no more revenue. Gartner call it agent sprawl, which is a polite name for it.

Gottlieb's other line in the same release is worth having on a wall. AI agents should not be viewed as a shortcut to sales productivity, because they're only as effective as the systems they operate within.

There's no silver bullet in there, and if there were, we'd all be rich. But the sequencing is genuinely simple. Fix the data, then buy the agent. Not the other way round, which is how almost everyone does it, because buying the agent is a decision you can make on a Tuesday and fixing the data is six weeks of unglamorous work nobody wants to own.

Nobody's Citing You

Now flip it round, because while you're pointing agents outward, your buyers are using the same technology to decide whether you make the shortlist.

Similarweb tracked how often AI assistants cite a source when they answer. In the US, citation rates went from 1.6% in June 2025 to 6.8% in May 2026. Big jump. But look at the split by category. Travel and hospitality sits around 23%. Automotive around 20%.

Professional services is under 4%.

So in the category where the buying decision rests almost entirely on trust and credibility, we're the least cited people on the internet. Two ways to read that, and I don't think they're mutually exclusive.

The generous read is that it's an open goal. Almost nobody in this industry is doing the work, so the bar to being the firm that gets named is currently low.

The less generous read, and the one I'd bet on, is that the models have nothing to cite. Go and read fifty consultancy websites. It's all "we partner with organisations to drive transformational outcomes across the value chain". There is no fact in there. No number, no date, no named client, no price, no method described in enough detail that anyone could check whether it works. A model cannot cite a claim that contains no information. Nor, incidentally, can a human buyer form a view from it.

The fix isn't a GEO tactic. It's putting real specifics in public. What the engagement costs. How long it takes. What it produced, with the figure. Who for, where they've agreed to be named. The firms that get cited in eighteen months will be the ones who published something checkable.

Touch And Timing

There's one more finding that cuts against everything I've just said, and I'd be doing the politician's answer if I skipped it.

Gartner also found 67% of B2B buyers prefer a rep-free experience, and 70% want a completely digital, self-service buying process. Which sounds like the end of selling.

Then, in the same body of research, 69% of those buyers said they go to a sales rep to validate the insights the AI gave them. And this is the part I find genuinely funny. 51% said they're more likely to hit misleading information from GenAI. 49% said the same about sales reps. So they trust neither of us, and they're using each of us to check the other one.

Worth being straight about the dates here. That buyer survey was published in May 2026 but fielded in August and September 2025, so it's older than the headline suggests. Take it as direction, not as this month's reading.

What it tells you is that the rep's job has changed rather than disappeared. You are no longer the source of the information. The buyer arrives having already read a summary of your market, your competitors and probably your pricing, some of it wrong. Your job is confirming what's right, correcting what isn't, and being the person who can answer the second question rather than the first. That's a harder job than pitching, not an easier one, and it needs someone who actually knows the subject rather than someone reading from a deck.

The Boring Conclusion

Two problems, one root. Agents that don't help because the data underneath them is a mess. Buyers who never shortlist you because there's nothing citable about you in public.

Both come down to the same thing, which is that most firms have never made their commercial reality legible. Not to a model, not to a buyer, and in a lot of cases not to themselves.

Start with the CRM you'd be embarrassed to open in front of a stranger, and the website that doesn't contain a single number. Neither of those is an AI project. That's rather the point.