Their Ads Stopped Working, So They Opened a Phone Line

A room of fashion and beauty founders sat down with Modern Retail and Glossy in September and, under rules that let them speak without their names attached, said the quiet thing. The channels they had used to find customers for a decade were not working the way they used to.

One in house marketer put it plainly. His company had always leaned on search advertising and display. Now, he said, it is very difficult to advertise on those channels because people increasingly use language models to find products. His answer was to put money back into in person contact. An apparel co founder in the same room said her brand had started throwing parties. A kids' fashion executive described store events with face painting.

None of that is a marketing theory. It is a group of operators describing what happened to their customer acquisition numbers, and the useful question for anyone running a small shop is whether the same thing is happening to theirs, and what the evidence says to do next.

What actually changed in the ad channels?

Nothing broke. The demand moved one step upstream, to a place where an ad cannot sit. When a shopper describes what they want to an assistant and gets three suggestions back, the auction that used to decide which shop they saw never runs.

This matters more for consideration than for intent. Somebody typing a brand name into a search box still sees ads and still clicks them. Somebody asking an assistant which showerhead works in a rental flat is having a different conversation, and the brands that appear in the answer were picked on the strength of what a machine could read about them, not on a bid.

The effect on a paid budget is indirect but real. The high intent searches remain and stay expensive. The broad discovery searches that used to fill the top of the funnel thin out, so the same spend buys fewer new names, and the cost of a first order drifts up while the reported click cost looks stable. We wrote about the mechanics of this in more detail when Google started reporting AI surfaces separately, in the piece on how Search Console now counts AI impressions rather than clicks.

Does AI traffic replace what search ads used to deliver?

Not yet in volume, but it is the highest quality traffic most retailers have. Adobe's July 2026 reading, reported by Digital Commerce 360, found AI referral traffic to US retail sites up 62% year over year, with those visitors generating 53% more revenue per visit and converting at a rate 60% higher than non AI traffic. It was the eleventh consecutive month of AI traffic beating everything else on conversion.

The engagement numbers underneath are consistent with people who already decided. Visitors from AI platforms spent 59% more time on site, bounced 33% less and added to cart 28% more often. That is what a pre qualified visitor looks like. The assistant did the comparison shopping before the click, so the session starts further down the funnel than a cold ad click ever does.

Read the growth rate rather than the headline, though. Against October 2024, Adobe puts the increase at 1,219%. In the first quarter of 2026 the year over year figure was in the hundreds of percent. By July it was 62%. The base is getting larger and the multiple is coming down, which is the normal shape of a channel maturing. It is not the shape of something about to replace paid search this quarter.

The row that should hold your attention is the second one, because it is the only one where the input is work you already own rather than money you do not have.

Why are brands putting people back on the phone?

Because being reachable has become rare enough to be worth advertising. Sproos, a showerhead brand aimed at renters, launched a staffed phone line this summer and takes about 30 calls a day on it, with co founder Benjamin Fix saying the point was to differentiate from all those AI robots.

Made In runs both a chatbot and a phone line, with staff including professional chefs taking video calls. Trade Coffee has had a line since 2018, staffs it with about six people plus seasonal contractors, and requires its agents to be former baristas. Its growth director says customer reviews name individual phone representatives as a reason people stay.

The consumer research points the same way. SurveyMonkey's December 2025 study of 2,017 US adults found that 79% strongly prefer interacting with a human over an AI agent. An earlier SurveyMonkey fielding put the preference at 90%, which looks like a contradiction until you read the questions: the 90% figure asks which people prefer for customer service generally, while the 79% figure asks about strong preference against an AI agent specifically. Both describe the same direction with different intensity, and neither says automation has no place.

Here is the part the coverage leaves out. Thirty calls a day is not free. At four minutes a call plus the interruption cost of context switching, that is somewhere between two and four hours of a person's day, every day. For a business with six support staff that is a rota decision. For one person who also packs the orders, it is most of an afternoon, and the honest answer is that you publish a number with hours attached and let voicemail handle the rest rather than pretending to be always on. If you are deciding what to automate and what to keep human, the sequencing matters more than the tooling, which is why we wrote about which support tickets are safe to automate first.

The money went to an enterprise answer

While independent brands were buying folding tables for events, investors were funding software. Profound, which sells visibility inside AI answers, raised $180 million at a $1.8 billion valuation less than seven months after a $96 million round, led by Sequoia and Kleiner Perkins, with revenue up threefold in six months.

Read the customer list before you read the valuation. Profound names Comcast, The Estee Lauder Companies and Walmart among more than 1,000 enterprise customers. This is a category being built for companies with a brand team and a procurement process. The problem it solves is real and it is the same problem the founders at that dinner described, but the version of it sold to a one person shop does not exist yet at a price a one person shop would pay.

That leaves independent sellers doing the unglamorous version by hand, which is mostly a data exercise rather than a marketing one.

What does machine readable actually mean for a product page?

It means a machine can extract the facts without guessing. Adobe found 39% of retailer homepages are not machine readable by language models, with apparel at 76% readable, electronics at 70% and grocery at 59%, and the gap between those categories is mostly a gap in how disciplined the product data is.

In practice that comes down to a short list. The price, the currency and the availability appear as text and as structured data rather than only inside an image. The variant that is actually in stock is distinguishable from the one that is not. Specifications that decide a purchase, the dimensions, the material, the fit, the compatibility, sit in a table or a list instead of being buried in a paragraph of brand voice. Delivery timing and the returns window are stated on the page rather than linked to a policy three clicks away. Every claim on the page is one a machine could quote without inventing anything.

None of that requires new software. It requires a catalogue where the fields are filled in consistently, which is why the shops that already sell on marketplaces tend to score better: the marketplace forced the discipline years ago. We went through the specific fields that matter in the piece on why AI shoppers convert better while product pages lag behind, and if you are building a storefront from scratch it is worth starting with a structure that already emits the data, which is the approach behind the AI store builder we ship.

What should a small shop change this quarter?

Start with the thing that costs nothing and compounds. Your product data is the input to every channel in the table above, including the paid ones, and fixing it improves the ad feed and the assistant answer at the same time.

The order that makes sense for a business without a marketing team looks like this. First, audit your twenty best selling products and fill in every field a buyer would ask about, in text. Second, check that price and stock state render without JavaScript, because that is where most extraction fails. Third, publish a contact route a person actually answers, with honest hours, and put it where a hesitant buyer sees it rather than three levels into a help centre. Fourth, start recording where new customers say they heard about you, because the analytics will not tell you and the assistant referral will often arrive with no referrer at all.

Only after that does it make sense to spend on reach. An event or a pop up is a reasonable use of money once you have somewhere to send the people who liked you, and a paid campaign is a reasonable use of money once the page it lands on answers the questions the ad raised.

How do you measure any of this without an analytics team?

Pick two numbers and watch them monthly, because the channel level reporting will lie to you here. The first is blended acquisition cost, which is every pound you spent on marketing in a month divided by the number of first time buyers that month. The second is the share of revenue from repeat customers.

Blended cost is the honest one precisely because it ignores attribution. If assistants are quietly sending you buyers, blended acquisition cost falls even though no report credits the source. If your paid discovery is thinning out, it rises even while the platform still reports a healthy cost per click, and that divergence between the platform number and the blended number is the single clearest signal that product discovery has moved somewhere your ads cannot follow.

The repeat share matters because it is what the in person strategy is actually buying. A party or a store event rarely pays back on the night. It pays back if the people who came order again, so the founders describing events as worthwhile are making a claim about retention, not about first orders. If your repeat share is flat six months after you started doing events, the events are a cost.

Two smaller habits help. Add an open text field at checkout asking how the customer found you, because assistant referrals often arrive with no referrer at all and a one line answer beats an empty report. And track conversion rate by landing page rather than sitewide, because the ecommerce traffic arriving from an assistant lands deep, usually on a product page, and averaging it with homepage sessions hides both the good news and the problem.

Retail brands with a media team are buying answer engine optimisation software to do a version of this at scale. Without one, the same intelligence comes from a spreadsheet with four columns and twelve rows a year, and it is enough to tell you whether structured product content is earning its keep.

What this does not mean

It does not mean paid search is finished. High intent queries still convert and still justify a bid, and a brand with no organic presence that switches off its ads will feel it within a week.

It does not mean every shop needs a phone line. It means being reachable is now a positioning choice rather than a cost centre, and a business that answers email in four hours has made the same choice more cheaply than one that staffs a line badly.

It also does not mean the assistants are sending you customers yet. For most independent sellers the volume is still small, and the case for preparing now is that the work required, a clean catalogue and honest product facts, is work that pays off in the channels you already have. If you want to know which assistants are actually in play, we tracked that in the piece on which assistants are sending buyers to shops.

The founders in that room were not predicting the end of digital marketing. They were describing a period where the cheap reach ran out before the replacement arrived, and filling the gap with the oldest thing in retail, which is a person who knows the product and is willing to talk about it. That is a reasonable answer. It is also a reminder that the only asset in this story a small merchant fully controls is the accuracy of what their own pages say.