Which channel to cut first when AI answers take your traffic, and the tool bill both answers quietly added

Asked which marketing channel to cut first if AI answers are taking the traffic, ChatGPT and Gemini agreed on the answer and disagreed on the bill. Both said do not cut a channel: cut the low intent informational content that answers simple questions and produces no qualified leads, and keep the channels that produce revenue. Both named top of funnel content as the first thing to stop paying for. Then the answers diverge. One priced three free diagnostic tools and told the reader to check whether AI referrals are being misclassified as direct traffic before moving any budget. The other recommended a five tool stack with prices, including an AI citation monitoring product at around 89 dollars a month. That vendor's own pricing page, read the next morning, renders no price at all.

What was asked, and what came back

One question, put to the ChatGPT and Gemini product surfaces through a data provider on 21 October 2026, at 22:21 and 22:38 UTC. Both ran a live web search.

• ChatGPT returned 63 citations, including six from Google support, four from Google developers, three from Google's own blog, six from one search trade publication, three from a Microsoft product, three from Microsoft Learn and two from a management consultancy. It opened by refusing the premise of the question and then gave three evidence points with their status attached.
• Gemini returned one citation. It named the cut immediately, top of funnel informational content, and then listed five products with prices and a reason for each.

The evidence one answer carried, and how it labelled it

These are the engine's claims and its attributions, recorded as it gave them on 21 October 2026. We did not verify the underlying studies, and they are named so a reader can.

• A 2026 field experiment reporting that Google AI Overviews reduced organic clicks by 38 per cent on searches where they appeared. The engine stated that this was a working paper and not peer reviewed, which is the single most responsible sentence in either answer.
• An analysis by Seer Interactive of 5.47 million queries, reporting that organic click through rates fell sharply during 2025 on AI Overview queries where the brand was not cited.
• A Semrush finding that AI referrals grew 66 per cent in 2025 and still accounted for under 0.15 per cent of visits in its dataset.

That third figure is the one to plan against, and the engine drew the right conclusion from it: a 0.15 per cent referral share is not a replacement for lost search clicks. Anybody proposing to cut a channel and move the budget into AI referral traffic is proposing to swap a measured channel for a rounding error. The reason to work on AI answers is not the referral click. It is being named and cited in the answer a buyer reads before they ever click anything.

Our own measurement puts a number on how concentrated that is. In 635 readings of US buying questions on 22 October 2026 we collected 19,793 citation URLs across 3,307 hosts, and half of all citations came from just 51 of those hosts. Being in the answer is a question of being on the small set of sites that get read, which is a different job from collecting referrals.

What to cut, in order, from both answers merged

1. Low intent informational content spend. Pages answering simple what is and how to questions that generate neither qualified leads nor assisted conversions. Both engines named this first and independently.
2. Paid search campaigns whose incremental acquisition cost exceeds your target. Incremental, not platform reported, which means a holdout or a geo test rather than the dashboard number.
3. Nothing else, yet. One engine was explicit: do not cut email, paid social or organic search across the board without evidence from your own attribution data. The decision rule it gave is contribution margin and qualified conversions rather than sessions, with a small reallocation measured over four to six weeks.

Before any of that, check one thing in your analytics: whether assistant referrals are landing in direct traffic. If they are, the channel you are about to cut may be carrying demand you cannot see. Google's own documentation is the place to settle what each report counts, and our note on what Search Console and Analytics each count on AI traffic works through it.

The five tool prices, checked the next morning

Read by us on 22 October 2026, each figure belonging to the URL it was read on rather than to the brand, from a machine on an Indian IP address.

Product and the price the engine gave	What the vendor page rendered
Ahrefs, Lite at 129 dollars a month and Standard at 249	Both match exactly. The page also renders Starter at 29, Advanced at 449 and Enterprise at 1,499 with an annual commitment, an AI brand product from 199 a month, and custom prompt packages at 50, 100 and 250 a month that the answer did not mention
Semrush, Pro at 139.95 and Guru at 249.95	Neither plan name is on the page. It renders Free, then SEO at 117.33 a month billed annually against 139 monthly, Starter at 165.17 against 199, Pro plus at 248.17 against 299, and Advanced as custom
Peec AI, starter plans from around 89 dollars a month	No dollar figure reached us. peec.ai/pricing rendered four plan names to our fetch route, Starter, Pro, Advanced and Enterprise, with Enterprise marked custom and annual, empty inclusion lists and feature labels including role based permissions and single sign on. A reading of the same page on 21 October did capture its per tier prompt and project limits, so the tiers are published and the 89 dollar figure is the part neither reading found
Google Analytics 4, free, with the enterprise tier from around 50,000 dollars a year	Not fetched. The free tier is the one both engines put first and neither charged for
HubSpot, Starter at 15 dollars per seat and Professional from 800 a month	Not verified. That vendor's pricing page returned a document to us whose price block did not render on the day

So of the three AI and SEO platform prices in that stack, one vendor's figures matched to the dollar, one was quoted under plan names the vendor no longer uses, and one was quoted at a dollar figure that neither of our two readings of that vendor's pricing page found on it.

The part we have to disclose, because it is about us

The second answer's fourth recommendation was a citation monitoring tool, to track whether a brand is cited inside AI answers rather than relying on rank tracking. That is the category we sell in, and the engine named a rival rather than us. It is not an isolated reading: across the 635 readings in this pass, rival vendors were named 1,053 times from a roster of 24, our own domain was named in 0 readings and cited in 0, and it appears in none of the 19,793 citation URLs we collected. We publish that number in the same place we publish our prices.

What AI Knows Us does is the measurement itself: the buying questions a customer would ask, put to ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews, with every brand named and every source cited recorded and dated. We buy no media, run no campaigns and cut no budgets, so this page is not a bid for the channel work underneath it.

Our own rates, verified on 22 October 2026, since this page prices four other vendors. In the US and the UK the ladder runs SEO Start at 149 dollars a month, Growth 299, Scale 599, SEO Scale 699, Inbound 1,099, Outbound 1,299, Agency 1,299 and Everything 1,799. Outside the US and the UK the first two tiers are 99 and 199, so the floor is a US and UK floor. SEO Start reads 0 questions on 0 engines, which makes Growth at 299 the cheapest plan that reads all five surfaces, and a Claude reading costs 3 credits where the others cost 1.

Where rivals are stronger, by name, from our own fetches that morning. peec.ai/pricing lists single sign on and role based permissions among its features. tryprofound.com publishes single sign on, SAML and SOC 2 with up to nine answer engines on its enterprise tier. ahrefs.com/pricing publishes a full ladder from 29 to 1,499 dollars a month plus an AI brand product from 199. Our own pricing document, 119,203 bytes read the same day, renders no single sign on, no SAML, no SOC 2, no ISO certification, no data processing agreement and no white label, and we read five surfaces rather than nine. It also contradicts itself for multi brand buyers, listing businesses covered as 1 1 1 1 1 1 10 1 across the plans while the row beneath reads more than one business, no, in every column.

The limits of this reading, stated plainly. It is one question per engine on one night, and four vendor pages fetched once the next morning, two of which did not render what we went to read. The three research findings quoted above are the engine's attributions to third parties rather than our measurements, and one of them is a working paper by the engine's own description.

Questions US marketers ask about cutting channels in an AI search year

Should I cut SEO because of AI Overviews

Neither engine said yes, and the referral arithmetic is why. Assistant referrals were reported at under 0.15 per cent of visits in one 2025 dataset, so there is nothing on the other side to catch the traffic you switch off. Cut the informational pages that produce no leads, keep the pages that answer buying questions, and measure contribution margin rather than sessions.

What should I cut first

Both engines gave the same first cut: spend on top of funnel informational content whose questions an assistant now answers on the results page. Then paid search campaigns whose incremental acquisition cost is above target, tested with a holdout rather than read off a platform dashboard.

How do I tell whether AI answers are actually costing me clicks

Segment your own query and landing page data year over year, and check whether your assistant referrals are being filed as direct traffic before you conclude anything about a channel. One engine put that check ahead of every budget decision, and it is free to run.

Do I need to buy an AI visibility tool to do this

Not to make the first cut. The diagnosis in both answers runs on free tools. A monitoring product earns its place when you need the same question set asked on a schedule across several engines, and at that point ask any vendor for a published price: of the three we checked on 22 October 2026, one publishes a full ladder, one published plan names and usage limits without a dollar figure reaching us, and one no longer uses the plan names it was quoted under.

How do I know what assistants say about my category right now

Ask the questions your buyers ask, with no brand name in them, on each assistant on the same day, and write down which businesses are named and which sources are cited. Repeat it a week later. In our own set the citation list was concentrated enough that half of 19,793 citations came from 51 hosts, so the useful output is the list of sites deciding your category, not a score.