Last week a Phoenix attorney told us he was passing on AI visibility because “the ROI isn’t established yet” and the space is “still early and evolving rapidly.” He’s investing in local SEO, reviews, and traditional client acquisition instead. Channels where the ROI is “more established.”
He’s not wrong about any of that. He’s also describing, almost word for word, what every arbitrage opportunity looks like before it closes.
This is a short essay about why the channels with “established ROI” are, by definition, the worst ROI channels you can buy into today, and why the channel everyone is uncertain about is the one with the highest expected return per dollar for the next 18 to 24 months.
What “established ROI” actually means in marketing
When a marketing channel has well established ROI, here is what has happened:
- Early adopters figured out it worked.
- Case studies got published.
- Agencies built service lines around it.
- Competitors poured budget in.
- Auction prices for attention rose to absorb most of the margin.
- Performance regressed to “fine, like everything else.”
This is the story of Google Ads from 2003 to 2015, Facebook Ads from 2012 to 2018, Amazon PPC from 2016 to 2021, TikTok creators from 2020 to 2024, and local SEO itself.
The keyword “personal injury lawyer Phoenix” cost a few dollars a click in 2008. Today it costs north of $200 per click on Google Ads, and ranking organically requires years of link building, content, and PR most local firms cannot sustain.
The reason it is expensive is not because it works better than it used to. It is because everyone agrees it works.
“Established ROI” is the receipt for the arbitrage other people already captured.
What “not yet established” looks like, historically
Every channel with established ROI was, at one point, the channel a smart, careful operator declined because “it’s too early.”
- Google Ads in 2003: “Nobody clicks on those text ads.”
- SEO in 2005: “Why would I write articles for a search engine?”
- Facebook Ads in 2012: “My customers are on Facebook for friends, not businesses.”
- Influencer marketing in 2016: “Those are just kids with cameras.”
- TikTok in 2020: “It’s a dance app.”
In each case, the people who moved first paid 1/10th to 1/100th of what the channel costs today, locked in distribution moats, and were already compounding when the agencies showed up to sell “TikTok strategy” at a markup.
The pattern is so consistent it has a name in finance: the alpha is in the channels that do not yet have a category.
Where AI visibility is right now
ChatGPT, Claude, Perplexity, Google AI Overviews, and Gemini are doing something Google did not do for its first decade. They are answering commercial questions directly, with named recommendations, before the user ever sees a SERP.
When someone asks ChatGPT “what is the best estate planning attorney in Phoenix for blended families?”, the model returns 2 to 4 specific firms with reasoning. The user does not see ten blue links. They see a shortlist. If your firm is not on it, you do not exist for that query.
The volume of these queries is, today, smaller than Google’s. We will not pretend otherwise. It is the single most important caveat. AI search is probably somewhere between 5% and 15% of the consideration journey for most local service categories right now, varying wildly by demographic and intent.
But three things are simultaneously true about that volume:
- It is growing fast. Every quarter, the share of high intent commercial queries that start (or end) in an LLM goes up. Not because the LLMs are better at search, but because users are migrating.
- It is the highest intent traffic on the internet. Someone who asks an LLM “best X in Y for Z situation” has already done the awareness work themselves. They are not browsing. They are buying. The closest analog is voice search circa 2018, except this audience converts.
- It is structurally cheaper to win than Google SERPs. Nobody is bidding against you. There is no auction. The “ranking factor” is whether the model has been exposed to enough structured, authoritative, AI readable content about you to confidently recommend you.
That third point is the arbitrage.
The arbitrage, stated plainly
In Google’s auction market, you pay for attention every time. You stop paying, you stop appearing.
In LLM citation markets, you pay once to be discoverable, and the citation compounds across millions of queries until something changes the model’s reference behavior. When ChatGPT decides your firm is one of the canonical answers for “estate planning attorney in Phoenix for blended families,” it gives that answer to every user who asks a variant of it. For as long as the underlying retrieval and training systems keep treating you as the answer.
There is no per click cost. There is no auction. There is a one time investment in being the kind of source LLMs prefer (structured, fact rich, schema marked, frequently updated, trusted by the broader web), and then a long tail of free, high intent placements.
This is the same shape as winning an SEO position in 2005, except SEO took fifteen years to mature into a commodity, and AI visibility appears to be on a 24 to 36 month curve.
The Phoenix attorney is correct that the ROI is “not yet established.” He is incorrect about what that implies.
“But local SEO and reviews are already working.” Yes, and that is why this stacks.
The strongest objection is not that AI visibility does not work. It is: I am already investing in local SEO, reviews, and reputation. Is this not the same thing?
It is not the same thing, but it is the foundation for the same thing, and that is why this argument actually cuts the opposite way.
LLMs decide who to cite based on signals very similar to what local SEO and reputation work produce:
- Structured, authoritative web pages
- Consistent NAP (name, address, phone) across the web
- A high volume of legitimate third party citations and reviews
- Topical depth on the specific niche (for example, “blended family estate planning” rather than generic “estate planning”)
- Clean schema and machine readable facts
A firm that has been investing in local SEO and reviews for two years is more primed to win AI visibility than a firm starting cold. They just do not know it yet, because nobody has translated those existing signals into AI readable form.
The 80/20 for these firms is not “start a new channel.” It is: take the foundation you already paid for, make it legible to the systems that are about to route high intent traffic, and capture the citations before competitors do.
That work (a structured content layer, schema, AI readable variants of the pages you already have, monitoring across the four or five models that matter, and a feedback loop on which prompts you appear in) is the entire job. It compounds on the local SEO investment. It does not replace it.
The firms with the strongest local foundations have the most to gain and the most to lose by waiting, because once a competitor down the street becomes the canonical answer to a query, you have to displace them rather than claim an empty seat.
What you give up by waiting 12 months
This is the part that is hard to feel until it is too late.
LLMs are not search engines. They have memory in two ways most people underestimate:
- Training data has lag. Whatever the model learned about your industry six months ago is still informing today’s answers. When a new model trains and your competitor is the cited authority, that lives in the weights for the next refresh cycle too.
- Retrieval indexes prefer the established. Even with live web search, models disproportionately cite sources they have cited before. Citations beget citations.
So the cost of waiting is not “we will start when it is bigger.” The cost of waiting is that when you start, you will be paying to displace an incumbent answer rather than become one. Anyone who has tried to outrank a well established SEO competitor on a money keyword knows what that costs.
The firms that establish citation positions in the next 12 months will, for most queries in their category, be paying past tense costs to capture present tense traffic for several years.
That is the arbitrage. It closes the moment everyone agrees it works.
How to think about the spend honestly
We are not going to tell you AI visibility replaces your local SEO budget, your review generation, your referral pipeline, or your paid acquisition. It does not. Anyone selling it that way is overpromising.
What it does, when done seriously, is:
- Capture a fast growing slice of the highest intent commercial queries in your category
- Turn the local SEO and reputation work you are already paying for into additional placements you were not previously getting credit for
- Establish citation positions that compound rather than expire
- Give you a measurable view of which models recommend you, for which prompts, against which competitors, so the work stops being a black box
The right way to budget it is not “is this better than my Google Ads spend?” It is “is the expected return on this dollar, given how few competitors are paying attention, higher than the marginal dollar I would spend on a channel that is already saturated?”
For the next 18 months, for most local service categories, the answer is unambiguously yes.
The honest closing pitch
“The space is early and evolving rapidly” is a true statement and a bad reason to wait. It is the same statement smart, careful operators used to pass on Google Ads in 2004, Facebook in 2012, and TikTok in 2020.
The cost of being early is some uncertainty and a smaller volume than the channel will eventually deliver. The cost of being late is paying market rates to displace incumbents on queries you could have owned for the cost of doing the work properly the first time.
If you are already investing in local SEO, reputation, and reviews, you have already done the expensive part. The remaining work, making those investments legible to the systems that are about to allocate a meaningful share of high intent commercial traffic, is the cheapest leverage available in marketing right now.
The best time to be discoverable to AI was twelve months ago. The second best time is before your competitor down the street figures it out.