Getting started
You're at $10M and never did SEO. Here's what actually works for AI search.
By Arnav Mukherjee, founder of TofuBofu · August 11, 2026
When we built the control group for our regional research, we pulled the firms ranked on a national industry list. Not companies we found, companies an independent publisher had vetted and ranked. Eighty-eight of them operated in the regions we had scanned. We checked how many the AI engines named when asked who the best providers in those regions were.
Sixty-six were named by nobody. Not once, by any engine. These are not struggling companies. They are profitable enough, established enough and good enough to appear on a published national ranking, and to an AI engine assembling a shortlist they did not exist.
If you have built a real business without ever buying an SEO retainer, that is the number that should interest you. It says the thing you have been quietly relying on, that being genuinely good and genuinely known eventually shows up, does not hold here. It also says something more useful, which is that the gap is not about quality. It is about what has been written down.
Your reputation is real and mostly private
A company that grew on referrals has its reputation stored in places no crawler can reach. It lives in the heads of clients who recommend you at industry events. In the sales call where someone says they have heard good things. In a decade of repeat business and renewals nobody ever wrote a case study about.
All of that is genuine evidence of quality. None of it is retrievable. An AI engine asked who a buyer should call assembles its answer from public text: your site, directories, review profiles, third-party roundups, forum threads, press. If the strongest evidence about you has never been published, the engine is not ignoring it. It cannot see it.
This is the specific failure mode of the successful, referral-grown company, and it is why the pattern catches people by surprise. The firms most likely to be invisible are frequently the ones who never needed marketing, because not needing it is exactly what stopped them writing anything down.
Where your reputation actually lives
Four things you have that a startup would pay for
The standard advice in this category is written for new brands, which is why it leads with schema and content calendars. You are not a new brand. You are sitting on the exact inputs that produce AI citations, and the reason to be optimistic is that publishing something true is far quicker than becoming true in the first place.
Customers who will go on record
A named client, quoted, with the problem and the result stated, is the single most quotable artifact you can publish. A startup cannot produce one because it has no clients yet, and a mid-sized competitor with a marketing team has usually produced six. In our matched crawls, published case studies were the axis where the firms getting recommended had the clearest lead in raw volume.
Outcomes with numbers attached
You know what you actually deliver: the response times you hit, the migration you completed in a weekend, the audit you have never failed. This is the specific language engines repeat when they describe a vendor. A page saying you deliver world-class service gives an engine nothing. A page saying what you did, for whom, and by how much, gives it a sentence it can quote.
A real position, not an aspirational one
You know exactly who you are good for and who you are wrong for, because you have lost enough deals to learn it. Most companies will not write that down, so their site describes a firm that serves everyone. Being explicit about the client you fit is what lets an engine match you to a qualified question, which is where a smaller name can still beat an incumbent.
The standing to ask for reviews and get them
Kevin Indig's analysis with G2 found roughly 10 percent more reviews associated with about 2 percent more AI citations. A company with hundreds of happy customers can build that in a quarter by asking. A startup cannot. Real reviews only, always.
Find out what the engines currently say about you
Run a free scan across six AI engines on your real buying questions. It takes a few minutes and tells you whether a decade of reputation made it into the answer.
Get your free auditTwo things you will be sold that you should decline
The first is a technical audit as the opening move. We crawled 14 firms that AI engines never named against the specific rivals the engines did name in the same regions. Organization markup sat on 86 percent of the invisible firms and 88 percent of the named ones, and more of the invisible firms served an llms.txt than the named ones did. The technical layer is real work and it is table stakes both groups had already cleared. If your site functions and has a sitemap, an audit will hand you a tidy list of small fixes and will not change who gets recommended.
The second is a thought-leadership content programme. Across 3,538 question-and-engine cells from real scans, brands were named on 5.4 percent of buying questions and on zero percent of the 602 awareness and informational cells. Not a low rate. Zero. Twelve posts about where your industry is heading will sit precisely in the band where no company gets named, and you will conclude twelve months later that this does not work.
Both of these get proposed because both are easy to scope and invoice. Neither is where your particular advantage is, and you have an advantage most buyers of this work do not.
The sequence that fits your situation
Order matters more than volume here, because every step after the first is cheaper once you know what the engines currently say.
Measure before you commission anything
Ask the five questions your buyers ask, across several engines, and save the answers with the date. You need this baseline before anyone starts work, because without it every future claim of improvement is unfalsifiable. Read who gets named instead of you and what the engine says about them. That sentence is your actual competitive gap, stated by the machine your buyers are asking.
Publish what is already true
Case studies with real client names and real numbers. The answers your sales team gives on every first call, written out as questions and answers. What you cost, or at least how you price. This is the highest-yield work available to an established firm and it requires no new capability, only the decision to write it down and the client permissions to name names.
Make the third-party record agree with itself
Directory entries, review profiles and association listings are where an engine goes for a second opinion. Where those records disagree about your name, location or category, you are measurably harder to verify than a rival whose entries line up. Then ask your happiest clients for reviews on the platforms that matter in your category, and keep asking. This is the step your position lets you do quickly.
Write the comparison pages nobody has written
Buyers choosing between you and a named rival are at the bottom of the funnel and the pages that serve them are frequently unwritten for smaller competitors. Start with the rivals your buyers actually raise, not the biggest name in the category, whose comparison queries are already contested by everyone.
Then, and only then, do the technical pass
Mark up the questions you published in step two as FAQ data. Add the schema. Ship the llms.txt. Doing this last is deliberate: markup is a wrapper, and by now you have something worth wrapping. Done first, it wraps a site that never answered the buyer's question, which is the situation most of the invisible firms in our crawl were already in.
Why the referral engine is the thing at risk
The objection I hear most from firms at this size is reasonable: our leads come from referrals, so why does any of this matter. The answer is that the referral is no longer the end of the process. It is the beginning of a verification step that did not used to exist.
G2's 2026 research found 51 percent of B2B buyers now begin vendor research on an AI chatbot, up from 29 percent, and 69 percent have switched vendor based on what AI told them. Forrester's 2026 study found 94 percent use AI somewhere in the buying process. A referred buyer who asks an engine about you and gets a thin answer, or worse a confident recommendation of somebody else, has downgraded you before you knew you were being considered.
That is the part that does not show up in your numbers. The meeting that never got booked leaves no trace in a pipeline. Revenue looks fine, close rates on the deals you do see look fine, and the top of the funnel quietly narrows. Search rank was always the floor and being named in the answer is a distinct layer on top of it. For a company that skipped the floor entirely and won anyway, the layer is where the whole game now is.
Frequently asked questions
We are profitable and well known in our market. Why would AI not name us?
Because the reputation you built is mostly private, and an engine can only retrieve what is public. Referrals, repeat business, relationships and word of mouth in your industry are all real, and none of them exist as retrievable text. Our own research found 66 of 88 firms ranked on a national industry list were named by no engine at all. Those are award-listed companies, vetted enough to make a published ranking, and they were invisible. Being established does not transfer on its own. It has to be written down somewhere an engine can read.
Should we start with a technical SEO audit?
It is rarely where the money is for a company at your stage. We crawled 14 firms that AI engines never named against the rivals the same engines did name in the same regions. Organization markup turned up on 86 percent of the firms nobody named, against 88 percent of the firms that were named, and more of the invisible firms served an llms.txt than the named ones did. The technical layer is table stakes that both groups had already cleared. If you have a functioning website and a sitemap, the audit will produce a tidy list of small fixes and will not change who gets recommended.
What do we have that a startup does not?
Four things, and they are the raw material AI answers are actually assembled from. Customers who will go on record, outcomes with real numbers attached, a genuine position in a category rather than an aspirational one, and enough goodwill to ask for reviews and get them. A startup has to manufacture all four and it takes years. You already have them. The work is publishing them, which is a much shorter path than creating them, and it is why an established firm can often move faster here than a funded newcomer.
Should we start a blog and publish thought leadership?
Not first, and not the kind most agencies will propose. Across 3,538 question-and-engine cells from real scans, brands were named on 5.4 percent of buying questions and on zero percent of the 602 awareness and informational cells. Industry-trends content sits exactly in the band where nobody gets named. The content that earns a mention is the content that answers the question a buyer asks when they are choosing: who is good at this, for a company like mine, in my market, and what does it cost.
How long does this take for an established company?
Faster than for a new brand, because the corroboration already exists in private and only has to be moved into public view. The first thirty days is mostly publishing things that are already true: case studies with named clients, the answers your sales team gives on every call, directory and review profiles that agree with each other. What you cannot shortcut is the retrieval and indexing lag afterwards, so measure on a monthly cadence against a fixed question set rather than judging it from one screenshot.
Do reviews actually matter, or is that just directory upselling?
They matter as evidence rather than as flattery, and this is one area where being established is a direct advantage. Kevin Indig's analysis with G2 found roughly 10 percent more reviews associated with about 2 percent more AI citations, and G2 holds about 22.4 percent share of voice for software. A company with hundreds of satisfied customers can produce that in a quarter. Real reviews only: fabricated ones violate the platforms' terms, are illegal in several markets, and get removed.
Our leads come from referrals. Why change anything?
Because the referral is increasingly being checked before it becomes a conversation. G2's 2026 research found 51 percent of B2B buyers now begin vendor research on an AI chatbot, up from 29 percent, and 69 percent have switched vendor based on what AI told them. A referred buyer who asks an engine about you and gets a vague or empty answer has quietly downgraded you before the first call. That failure is invisible in your pipeline, because the meeting that did not get booked never appears in it.
Sources and further reading
- TofuBofu MSP research 2026: the control group drawn from a national industry ranking, where 66 of 88 listed firms were named by no engine, and 139 of 228 directory-listed firms likewise.
- Our measurement of 3,538 question-and-engine cells: 5.4 percent on buying questions, zero percent across 602 awareness cells.
- G2 2026 B2B buyer research: 51 percent begin vendor research on an AI chatbot, up from 29 percent, and 69 percent switched vendor based on AI.
- Forrester B2B Buying Study 2026: 94 percent of buyers use AI somewhere in the buying process.
- Kevin Indig with G2: G2 holds about 22.4 percent share of voice for software, and roughly 10 percent more reviews is associated with about 2 percent more AI citations.