TL;DR: The sceptic is right about most of it. The metrics are weak, the attribution is broken, the timelines are long and the channel is small. What survives that is a narrower argument: the work is largely things you should do anyway, the measurement is cheap, and the cost of being absent is asymmetric. Fund it at a size that does not require anyone to believe a forecast.
Key Takeaways
Published figures put AI referrals at around one percent of sessions, converting at several times the organic rate.
Most of the work is shared with search, so the incremental cost is small.
The measurement is a spreadsheet, not a platform, for the first year.
Do not argue from urgency. First-mover claims in this field have nothing behind them.
Sometimes the answer is no, and the page says when.
Propose a quarter, not a programme. That is the proportionate ask.
Anchor targets assume your CMS generates slug IDs from heading text. Confirm on staging and adjust the hrefs if your platform slugifies differently.
The sceptic is mostly right
Taking the objections seriously, because most of them hold.
The metrics are weak
True. There is no agreed definition of a citation rate, no published benchmark with a stated method, and every vendor score is a composite you cannot reproduce. Figures carry ranges wide enough that small changes are indistinguishable from noise.
Anyone who says otherwise is selling a dashboard.
The attribution is broken
Also true. Many assistants pass no referrer, influence without a click leaves no trace at all, and the visitor who arrives by search a week after reading about you in an AI answer is recorded as organic.
You cannot close this gap, only acknowledge it.
The timelines are long
True. Corrections to cited sources move in weeks; content moves in quarters; earned coverage moves slower still. A defensible answer on whether the programme worked arrives around month twelve.
The channel is small
True, currently. Published analysis across a large set of domains put AI referrals at roughly one percent of sessions.
What the sceptic gets wrong
Only one thing: concluding that because the measurement is weak and the channel is small, the work is not worth doing. That conclusion would hold if the work were expensive and separate.
It is mostly neither, which is the argument the rest of this page makes.
Why starting here matters
Because a business case built on the hype gets dismantled in the first meeting by someone who has read one critical article. A case that concedes the weak parts first is harder to dismiss.
What the channel actually is
The figures worth knowing, with their limits stated.
What has been measured
Finding | Source | Date |
|---|---|---|
AI referrals around 1% of sessions across 13,770 domains | Conductor | Nov 2025 |
AI-referred visitors converting at several times the organic rate | Semrush, 500+ topics | Jun 2025 |
ChatGPT recommending around 1.2% of local businesses tested, against 35.9% appearing in Google's local 3-pack | SOCi Local Visibility Index | Jan 2026 |
All three are over a year old and should be re-verified. They are also the best available, which is itself informative about the state of the evidence.
How to read them
Small, high quality, growing. That is the honest three-word summary and it supports a modest allocation rather than a reorganisation.
The conversion point is the interesting one
If AI-referred visitors convert at several times the organic rate, a one percent traffic share is worth more than one percent of your pipeline. That is the strongest single number in the field.
It is also one study, and the mechanism is plausible: someone arriving after a system described you as a fit has been pre-qualified in a way a search click has not.
What nobody has measured
Influence without a click. Someone reads an AI answer naming three options, does not click, and searches for one of them later. That path is invisible to everyone and is plausibly larger than the measured referral share.
That cuts both ways in an argument: it means the channel may be bigger than one percent, and it means you cannot prove it.
The honest framing for a sceptic
The measured channel is small. The unmeasurable part may not be. The work is cheap. That combination is what the case rests on, rather than any forecast.
The argument that survives
Three points, all of which hold even if the channel stays at one percent.
1. The work is mostly shared
Activity | Serves AI search | Serves search |
|---|---|---|
Crawler access and rendering | Yes | Yes |
Answering questions directly | Yes | Yes |
Consistent facts across sources | Yes | Yes |
Accurate listings and profiles | Yes | Yes |
Earned coverage | Yes | Yes |
Nearly everything in the first column is already justified by the second. The incremental ask is measurement and a prioritisation change, not a new programme.
That single table is the strongest thing in any internal case, because it reframes the question from should we invest in this to should we spend three hours a month checking.
2. The errors are worth fixing regardless
The measurement reliably turns up wrong statements about your product in sources that are being used: stale pricing, removed limitations, capabilities you now have described as absent.
Those are costing you in search and in sales conversations too. Finding them is worth the exercise even if no AI system existed.
3. The downside is asymmetric
The cost of doing this at the scale described is a few hours a month and some editing. The cost of discovering in two years that your documentation has been blocked to AI crawlers since a replatform is a large content investment that produced nothing.
That asymmetry justifies the check, not the campaign.
What this argument does not claim
That the channel will grow, that early movers gain lasting advantage, or that competitors are pulling ahead. None of those is established.
Why it is more persuasive
Because it survives the sceptic's objections rather than contradicting them. You are agreeing the channel is small and arguing the work is smaller.
Building the case
What to put in front of whoever approves it, in the order that works.
Open with the concession
This channel is currently small, the measurement is imperfect, and I cannot promise attribution. Starting there buys credibility for everything after it, and it pre-empts the objection rather than inviting it.
Then the shared-work table
The one from the previous section. It moves the conversation from a new investment to an adjustment of existing work, which is a far smaller decision.
Then the specific finding
Run the measurement before the meeting. Walk in with one real example: here is an assistant telling a buyer we do not support something we shipped last year, and here is the page it came from.
That single concrete item does more than any argument about market shifts.
The ask
Element | Size |
|---|---|
One person's time | Three to six hours a month |
Tooling | None initially. A spreadsheet |
Content | Reprioritised, not additional |
Technical | An hour of platform time quarterly |
Review point | One quarter |
What to promise
Findings and corrections in the first quarter. A direction by the third. A defensible answer on whether to continue at twelve months.
Nothing about revenue, and no target rate.
What this framing achieves
It makes approval cheap. Nobody has to believe a forecast to approve three hours a month, and the programme gets its first quarter without anyone staking their credibility on the category.
The follow-up that matters
Report the findings honestly at the review point, including if nothing moved. That is what buys the second quarter and it is what most programmes skip in favour of a chart.
Arguments not to use
Five that appear in every piece on this topic and will cost you the room.
Argument | Why it fails |
|---|---|
Early movers gain lasting advantage | Nothing establishes this. Positions move in both directions and no lock-in mechanism has been demonstrated |
Brands without AI visibility become invisible | The measured channel is around one percent of sessions. Invisible overstates it considerably |
Your competitors are already doing this | Usually unverifiable, and the person you are pitching may know it is not true |
AI is replacing search | Search still drives most sessions for most sites, and overclaiming here discredits the rest |
Citation momentum compounds and becomes unassailable | A mechanism that sounds plausible and has nothing behind it |
Why these are tempting
They create urgency, and urgency gets budget. They are also the first things a sceptical executive will test, and the whole case falls with them.
The first-mover claim specifically
It is the most repeated and the least supported. Corroboration accumulates slowly, which is true, and that is not the same as a durable advantage that competitors cannot overcome.
If anything, the slowness cuts the other way: a competitor starting two years later is two years behind on accumulation, not permanently excluded.
What to say instead of urgency
This is cheap, it overlaps with work we already fund, and it surfaces errors that are costing us elsewhere. Let us do a quarter and look at what it finds.
That is an argument nobody needs to be convinced of to approve.
The credibility trade
Urgency arguments get a faster yes and a worse second meeting. The modest case gets a slower yes that survives contact with results.
One more to avoid
Any statistic you cannot source. This field is full of them, and quoting one that gets checked ends the conversation permanently.
When the answer is no
Some businesses should not do this yet, and saying so is what makes the rest credible.
The cases
Situation | Why to wait |
|---|---|
Your buyers do not use assistants | Ask them. If nobody mentions it, the channel is not there for you |
Your fundamentals are broken | Unindexed pages and stale listings matter more, and fixing them serves both |
Nobody has three hours a month | A programme nobody runs produces nothing and costs credibility |
You sell through a channel exclusively | Your partners' visibility may matter more than yours |
You are in a crisis | Everything non-essential waits, and this is non-essential |
The sector question
For a local trade business, research found assistants recommending a very small fraction of locations tested compared with Google's local results. That bounds the opportunity honestly and makes an afternoon of profile work the right size of response.
For a B2B software company whose buyers research in assistants before shortlisting, the case is stronger. The difference is real and worth establishing before committing anything.
How to find out
Ask ten customers how they researched before they contacted you, and add the question to your intake form. A quarter of those answers settles the question with your own data.
That costs nothing and it beats every published statistic for your specific situation.
The honest position for most businesses
Somewhere in between. Worth the measurement and the error-fixing, not worth a dedicated budget line yet.
What waiting costs
Less than the urgency arguments claim. Corroboration accumulates slowly in both directions, so a team starting a year later starts a year behind rather than permanently excluded.
When to revisit
Annually, or when buyers start mentioning assistants unprompted. That second signal is the one that actually matters.
What to propose
A quarter, sized so that approving it requires no belief in the category.
The proposal
Month one. Check crawler access and rendering. Build twenty prompts from real sales calls. Run the first measurement. Record a baseline with its ranges.
Month two. Fix every factual error found, starting with your own pages and listings. Approach cited sources where you are described wrongly.
Month three. Rewrite the openings of ten important pages. Run the measurement again. Write up what was found.
What it costs
Resource | Amount |
|---|---|
One person | Three to six hours a month |
Platform or engineering | An hour, once |
Content | Ten rewrites, an hour each |
Software | None |
What you report at the end
Access status, the baseline with ranges, the errors found and fixed, the pages rewritten, the competitors appearing instead of you, and a recommendation on whether to continue.
Not a rate change, which one quarter cannot show.
The decision at the review
Three outcomes, all legitimate: continue at the same size, expand because the findings were substantial, or stop because they were not. Naming all three in advance makes the proposal credible.
Most teams find enough in the first quarter to continue, and the ones that do not have learned something worth knowing cheaply.
Why a quarter rather than a year
Because a year is a programme and a quarter is an experiment. The ask should match the evidence, and the evidence currently supports an experiment.
The thing that makes it work
Doing the measurement before the meeting, so you arrive with a real finding rather than a proposal. That single preparation step changes the conversation entirely.
Six mistakes
1. Arguing from urgency. First-mover advantage, competitors pulling ahead, becoming invisible. None is established, all get tested in the room, and the case falls with them.
2. Promising attribution. Many assistants pass no referrer and influence without a click leaves no trace. Promising a revenue number guarantees a conversation you lose later.
3. Asking for a programme. A quarter at three hours a month gets approved without anyone believing a forecast. A budget line requires conviction nobody has yet.
4. Skipping the measurement before the meeting. Walking in with a real wrong statement about your product beats every argument about market shifts.
5. Quoting unsourced statistics. This field is full of them. One that gets checked ends your credibility on the topic permanently.
6. Not naming when the answer is no. A case that admits some businesses should wait is far more persuasive than one that says everyone must act.
The pattern
Most come from overselling a modest, sensible activity. The honest version is easier to approve, because it asks for less and survives scrutiny.
The one to do first
Run twenty prompts across three assistants before you ask anyone for anything.
You will find something specific and wrong about your own company, and that finding is worth more in the meeting than this entire page.
FAQs
Is GEO just a buzzword?
The term is contested and the underlying activity is real but smaller than claimed. Published analysis puts AI referrals at around one percent of sessions, converting at several times the organic rate.
Can I prove ROI on AI search work?
Not cleanly. Many assistants pass no referrer and influence without a click leaves no trace. You can show errors corrected, questions you now appear for, and a trend after three measurement cycles.
How big is the channel actually?
Small and high quality on current evidence: roughly one percent of sessions across a large domain sample, with better conversion. The unmeasurable influence without a click may be larger and cannot be proven.
What is the strongest argument internally?
That most of the work is already justified by search: crawler access, direct answers, consistent facts, earned coverage. The incremental ask is measurement and prioritisation, not a new programme.
What arguments should I avoid?
First-mover advantage, competitors pulling ahead, becoming invisible, AI replacing search. None is established and each will be tested by the person you are pitching.
How much should I ask for?
Three to six hours a month for one quarter, no software, with content reprioritised rather than added. That is approvable without anyone believing a forecast.
When should we not do this?
When your buyers do not use assistants, when your search fundamentals are broken, when nobody genuinely has the hours, or when you are mid-crisis. Ask ten customers how they researched you.
What should I show at the review?
Access status, the baseline with ranges, errors found and fixed, pages rewritten, competitors appearing instead of you, and an honest recommendation including the option to stop.
Measure before you ask
Run twenty prompts across three assistants before proposing anything. You will find something specific and wrong about your own company.
That finding is worth more in the meeting than any argument about where search is going, because it is checkable, it is yours, and it is already costing you.
The proposal that gets approved
Concede the weak parts first: small channel, imperfect measurement, no clean attribution.
Show the shared-work table. Most of this is already justified by search.
Lead with the real finding you measured beforehand.
Ask for a quarter, three to six hours a month, no software.
Promise findings and corrections, never a rate or a revenue number.
Name the stop condition, which is what makes the rest credible.
If the gap turns out to be production
It usually does, once the measurement names the questions you lose. Blazly SEO covers research, planning, drafting and direct publishing from about $39 a month.
Summary
The sceptical case against GEO is mostly correct: the metrics have no agreed definitions, attribution is structurally broken because many assistants pass no referrer, the timelines run to quarters, and published analysis puts AI referrals at around one percent of sessions. What survives all of that is narrower and still worth acting on. Nearly every activity involved is already justified by search, so the incremental ask is measurement and prioritisation rather than a new programme. The measurement reliably surfaces wrong statements about your product in sources being used, which cost you regardless of channel. And the downside of not checking is asymmetric, since a crawler block discovered two years late invalidates a large content investment. Propose a quarter, concede the weak parts first, and name the condition under which you would stop.