Your next customer may never look at your website. A machine will read it, compress it, and hand a human three options. You are no longer charming a person in a chair – you are being quoted by something with no patience and no taste for adjectives.

PART ONE: HOW OFTEN IS THIS ACTUALLY HAPPENING?

1. Today: 45% of consumers already use AI in the buying journey (IBM, Jan 2026) – 41% research, 33% reviews, 31% deal-hunting.

2. Next season: 60% of shoppers expect to use an agent within 12 months (Kearney). By 2030, half of online shoppers will, carrying ~25% of their spend.

3. The traffic already arrived: AI-referred retail traffic grew 393% year over year in Q1 2026 (Adobe); Shopify orders from AI search rose 13x.

4. And it converts: AI-referred visitors convert ~42% better than traditional search – +37% revenue per visit, +87% time on site.

5. B2B is the bigger stadium: Gartner expects 90% of B2B purchasing to route through agents by 2028: $15 trillion, machine to machine.

6. The trust gap is the strategy: 65% of Americans trust AI to compare prices; only 14% let it buy unsupervised. The agent makes the shortlist, the human signs it.

7. You can’t see it happening: About 70.6% of AI referral sessions log in GA4 as “direct” – you are read far more than your dashboard shows.

PART TWO: THE WORLD CUP PART

8. Nobody gets signed off a highlight reel. An agent scouts you like a federation scouts a striker: minutes played, conversion rate, injury history. “Award-winning” is not a stat.

9. Semi-automated offside doesn’t hear the crowd. It reads coordinates. Your hero video and parallax scroll are crowd noise. JSON-LD is the coordinate.

10. It’s a group stage, not a final. Agents shortlist three to five. FAQ schema makes content ~3.2x likelier to surface; full Product + Review schema hits ~61.7% citation.

PART THREE: WHO DOES WHAT ON MONDAY MORNING

11. Designers: beauty that can’t be parsed is decoration. Real headings, real text – 44.2% of LLM citations come from the first 30% of a page. Answer first, then be gorgeous.

12. Developers: JSON-LD for Organization, Product, Offer, FAQ. Server-render the facts. Allow GPTBot, ClaudeBot and PerplexityBot – a quarter of the top 1,000 sites block them. Ship llms.txt: hours, not weeks.

13. Marketers: trade adjectives for numbers. Princeton’s GEO research measured up to 40% visibility gains from citing statistics and sources. About 86% of AI citations trace to brand-controlled pages.

14. Owners: pick one thing you can prove you are best at, publish the proof with a date on it, keep the date fresh. Vagueness is now a pricing disadvantage.

The referee is a machine now. Argue with it, or hand it clean footage and let it put your name on the scoreboard. 88% of marketers have already started; the other 12% are about to learn what relegation feels like.

Sixteen ways to put your Google reviews back to work

Everybody knows the line from the baseball movie. Build it and they will come. That is the biggest lie in small business. You built it, you do good work, and a stranger comparing four names on a phone at 10:47 on a Tuesday night has no idea you exist.

Here is the part that stings. Your customers already wrote your ad for you. They typed it into a little box on Google, hit post, and went to bed. It got read by nobody and then it sank.

This is not a soft idea. Nielsen has been reporting for years that people trust other people way more than they trust companies. Edelman’s yearly trust survey keeps showing the same slide, where the big official voices lose ground and regular voices pick it up. Researchers at Harvard who studied restaurant ratings found that one extra star moved sales enough to decide whether a place kept its doors open. You can borrow that trust. You just have to carry it somewhere people will actually see it.

1. Use her sentence. A heating company pays real money for a headline like Comfort You Can Count On. That same company has a review from a woman who wrote that the tech put covers over his boots without being asked and told her the price before he touched a single screw. One of those two sentences was written by somebody who was actually in the house. Guess which one belongs on the homepage.

2. Answer the fear first. Every business has one worry nobody says out loud. For movers it is not price, it is watching strangers carry your grandmother’s dresser down a stairwell. Somewhere out there is a review where a customer mentions that nothing broke and then spends the rest of the paragraph on the fact that nobody made her feel rushed. That second part is the sale.

3. Three stories per proposal. Roofers, remodelers, IT firms, and commercial cleaners all send documents that end with a price and a signature line, which is a cold way to finish a pitch. Put three short customer stories on the last page instead. Swap them by job type so the church renovation prospect reads about a church.

4. Praise at the register. The invoice. The booking page. The confirmation email. The hold music. The sign by the front desk. A dentist’s payment page is a moment of quiet dread, and one line from a patient who admits she put the appointment off for six years does more work there than any certification badge.

5. One review, one post. You will never stare at an empty content calendar again. A dad writes that his kid finally stopped covering his mouth in photos. That is an orthodontist’s whole quarter of marketing, sitting in one line, written by somebody with nothing to gain.

6. Retell, do not screenshot. A cropped screenshot slapped on a stock background is lazy and everybody can smell it. Tell the story in your own words, then say what happened next. Thirty words of context turns a testimonial back into a story, and stories are the only thing people share.

7. Praise recruits people. Trades, dental groups, vet clinics, and home care agencies are all fighting over the same shrinking pile of good workers, and every one of those job ads reads the same. Drop three real customer stories into the posting. Nobody wants to work at a place nobody thanks.

8. Read them out loud. Monday meeting, out loud, using the name of the person mentioned. There is a review on some vet clinic’s page where the writer barely mentions the diagnosis and spends four sentences on the doctor sitting down on the floor next to the dog. That hits somewhere a performance review never will.

9. People tell you what stuck. Stack up a hundred reviews and count what keeps coming up. Nobody writes about your equipment. If the auto shop keeps hearing about the mechanic who walked customers out to the bay and showed them the busted part, that is not a nice touch, that is the product. Hire for it. Train for it. Put it in the ad.

10. Paragraphs beat stars. A four star review with six sentences of detail sells harder than a five star that says great job. Stop sorting your best material by rating and start sorting it by story. The one people believe is the one that mentions a hallway, a time of day, or a name.

11. Old praise reads dead. A wall of glowing reviews that stops in 2021 tells a stranger the good crew already left. Fresh ones are their own kind of proof, separate from what they say. Somebody comparing four names on a phone at eleven at night is checking whether you are even still open.

12. Ask at the peak. Most companies send the automatic text three days later, which is about three days after the feeling wore off. Ask when the boat dealer hands over the keys, when the crew pulls the last fan out of a flooded basement, when the surgery goes fine. Tell them straight that what they write helps the next scared person pick. You will get paragraphs instead of stars.

13. Give it a face. Thirty seconds, one phone, somebody on your team reading a customer’s story out loud. Better, ask the customer to say it again on camera. Med spas, wedding venues, and injury firms live on this, but a storage facility can run the same play.

14. Customers write better ads. Take the strongest line a customer gave you, clean it up, and run it against whatever your agency wrote. It will be shorter, stranger, and more specific, and it will usually win, because it came from somebody who was not being paid to like you.

15. Your reply is content. Answer all of them, the good ones too, because your response is public and permanent and read by strangers. A funeral home writing back to a family a year later and remembering the son’s name is not doing customer service. It is publishing.

16. The bad one sells too. Nobody believes a perfect profile. The person reading your one star review is not checking whether you are flawless, she is checking who you turn into when something goes wrong. Answer it like fifty people are watching, because fifty people are.

Somewhere in your company there is a person who thinks repeating nice things about yourself is tacky. It is not. Your customers said it, they meant it, and sitting on it does not make you humble. It makes you quiet while a competitor doing worse work does all the talking.

There is an old Christmas movie where a department store Santa tells a mother she can get the toy cheaper across the street. Everyone in the room figures he just cost the store a sale. Instead there is a line out the door, because he told the truth and she told everybody she knew. That is the whole thing right there. Say it out loud. They already wrote the words for you.

Book Two of the Iliad stops the war cold so Homer can list the ships. Who sailed, from where, how many hulls, under whose command. It is the passage every reader skips, and it is the passage scholars have been citing for three thousand years, because it is the one place the poem quits singing and starts counting. The war reads as real because somebody did inventory.

That is the whole argument. The rest of this is just me making it inconvenient for you.

1. The research keeps finding the same two levers

The academic work on generative engines did something refreshingly dull. Researchers took source pages, rewrote them nine different ways, pushed them through the systems that write answers, and measured which rewrites got pulled into those answers more often. Keyword tricks did essentially nothing. Two edits did nearly all of the work: real statistics with sources, and quotes from named humans with credentials. Visibility moved by double digits, and it moved hardest in the crowded categories where every vendor already sounds like every other vendor.

The reason is not mysterious. A model composing an answer needs somewhere to put specificity. It has four hundred interchangeable ways to say your industry is changing fast. It has exactly one way to say that dwell time at grocery distribution centers in the Southeast runs 2.7 hours against a 45-minute appointment window – and that sentence has your name welded to it.

2. Most “proprietary” data is just unopened

Every operator I meet has a folder. Warranty returns by SKU. Denial codes by payer. Dwell times by receiver. Reorder intervals going back to the Obama administration. The reflex is to call it proprietary. Usually it isn’t proprietary, it’s just unpublished. Proprietary means a rival could beat you with it. Unpublished means nobody can use it at all, including you, while it quietly rots in a drive nobody has opened since the last CRM migration.

There is a parable about a servant who buries the money he was given. He is not rebuked for losing it. He is rebuked for making it inert. Sitting on data is not a neutral act.

3. Go dig. Here is where to put the shovel.

Stop hunting for somebody else’s study to cite and go count something only you can count. A few I would pay money to read:

• A freight brokerage already knows detention hours by receiver type and region. Nobody has ever published what an average dock actually costs a driver. That figure would be quoted in logistics answers for the next five years.
• A dental group has fifteen years of charts. What share of patients who skip two consecutive cleanings need a crown within five years? The whole profession is guessing. You have the answer sitting in your practice management system.
• A commercial insurance broker can say which kitchen layouts generate fryer burn claims, and at what rate per thousand covers served.
• An equipment dealer in the Corn Belt can put bushels on each day of planting delay, banded by soil temperature. Growers argue about this at the co-op every spring with nothing but folklore.
• A medical billing shop knows denial rates by payer and code. Better yet, it knows what share of denials come back paid on first refile – a number that is worth a keynote.
• A regional staffing firm knows time-to-fill by metro and role. Every CFO in the country is currently estimating that number badly.
• A residential mover knows breakage by packing method. Publish it and you have written the definitive citation on whether dish barrels are worth the upcharge.

None of that needs a data science function. It needs one person with database access and a free afternoon.

4. Then put a human being’s name on it

The second lever is a person, not a brand voice. A named human with credentials – a title, a tenure, twenty-two years on loading docks, licensed in four states, board-certified in something. These systems learned from a corpus in which credentials predict reliability, so they reach for the sentence somebody is accountable for. An unattributed claim from a company is marketing. The same claim from your named VP of operations is testimony. Identical words, completely different gravity.

5. The giveaway is the distribution

The Rosetta Stone is a tax decree. Bureaucratic housekeeping about temple revenues, of no importance to anyone. It survived because it was carved in public, in three scripts, in a place where strangers could read it. The confidential decrees of that era are gone. The one they gave away unlocked a language.

You will not get distribution by parking the good stuff behind a form. A machine cannot fill out your form, which means every gated PDF is now invisible to the thing your buyer is asking. And this race has a very short lead: the first credible number on a topic becomes the anchor, gets restated, gets cited by people citing the people who cited you, and hardens into the default answer. Second place gets summarized as “some estimates suggest.”

6. Legal will say no. Here is the yes.

The answer is almost always the same three moves: aggregate, lag, anonymize. Nothing below two hundred observations. Ninety days delayed. No client names, no geography tighter than a metro. State your method plainly – sample size, window, what you excluded and why. That is a defensible dataset, and frankly it is more rigor than most of what your industry currently passes around as a benchmark.

7. Publish the dog that didn’t bark

Holmes cracks one case on the fact that the dog stayed quiet. Negative findings are findings. The retention tactic that moved nothing. The material that failed in salt air. The segment that never converted no matter what you spent. Almost nobody publishes these, which is precisely why they get quoted – a system trying to give a balanced answer has almost nothing to reach for on the “it didn’t work” side of the question.

Last thing

Borges imagined a library containing every possible book and made the quiet point that total information is indistinguishable from noise. That is the corpus these machines are reading now. Infinite shelves, mostly interchangeable. The way out is not to write more. It is to be the one shelf with a number on the spine, a name underneath it, and a date.

Go find your ships. Count them. Publish the list.

Why directories, associations and conference programs quietly decide whether an AI ever says your name

For twenty years the game was ranking. Ten blue links, everybody elbowing for the top three, and an entire industry built on the theory that page two of a search result is where URLs go to die. That game is not over, but a second one has started next to it. Somebody types a question into a chatbot and gets back three names and a sentence apiece. No links to scroll. No page two. Just a short list, handed over with the serene confidence of a machine that has never once been unsure of anything.

Shakespeare had Juliet ask what’s in a name, and decided the answer was: not much, a rose smells the same either way. Respectfully, Juliet was not doing business development in 2026. Right now a name is a retrieval key, and if a machine cannot find yours in enough places, written down by enough people who are not you, then as far as that machine is concerned you are a rumor.

The fix is unglamorous and very old-fashioned: join things. Directories, member rosters, trade associations, industry clubs, licensing boards, alumni lists, award shortlists, speaker programs. Show up at conferences that print your name and your company somewhere a crawler can reach. Here are ten reasons that works, and what these systems are actually doing behind the curtain.

1. Corroboration beats claims. Your website says you are a leading advisor in your field. Of course it does; nobody’s About page says otherwise. A retrieval system treats your own site as a single witness with an obvious motive. A member roster, an association directory and a conference agenda are three additional witnesses with nothing to gain. Journalists call this sourcing. Machines do the same thing, faster and with less coffee.

2. You become an entity. Before answering anything, these systems try to resolve a name into a thing – a node with attributes, sitting in a knowledge graph. This is the same architecture a major search engine shipped over a decade ago, now doing much heavier lifting. Repeated, consistent listings of your name, title, company, city and specialty are what turn a string of characters into a node. Inconsistent listings turn you into three half-people.

3. Retrieval, not memory. The dominant technique in the research literature is retrieval-augmented generation: the model does not recall you from training, it goes and fetches supporting passages at the moment of the question, then writes around them. That means the unit of visibility is the passage, not the website. Directory entries are short, structured, fact-dense and self-contained – close to a perfect passage. Your 1,800-word narrative about the founding journey is close to a perfect nap.

4. You are your neighbors. Language models learn meaning from proximity; words that keep appearing near each other get treated as related. Sitting inside a specialist association’s member list drops your name into a dense cloud of the exact vocabulary your buyers use. One firm I watched went from invisible to routinely named in its category without publishing a single new blog post – it just got itself into four rosters where the category language already lived.

5. Structure is a shortcut. Directories publish in tidy, repeating templates, usually with schema markup underneath. Clean fields, predictable labels, machine-readable. Parsing your listing costs a system almost nothing; parsing your hero image and animated scroll effects costs it a lot. Being easy to read is an underrated competitive advantage, in software as in people.

6. Conferences mint citations. A conference is a citation factory that also serves bad coffee. Speaker pages, session agendas, sponsor lists, press releases, post-event recaps, someone’s enthusiastic write-up – all timestamped, all third-party, all indexable, all naming you and your company in the same breath as your topic. One panel can seed a dozen independent mentions. Even the badge scan is optional; the program is the point.

7. Recency carries weight. Retrieval systems tilt toward fresh material, partly to avoid confidently reporting things that stopped being true. A directory entry from 2019 with a dead phone number is a fossil, and worse, a contradiction the system now has to resolve. Current listings say you still exist. That is a lower bar than it sounds, and plenty of people are failing it.

8. The long tail wins. Nobody asks a chatbot for the best consultant. They ask who handles compliance readiness for early-stage payment startups in the Midwest, and they expect a real answer. Niche directories are built entirely out of that kind of specificity – the sub-sub-category, the credential, the region – which is precisely the language a narrow question is fishing for. Broad visibility is expensive. Narrow visibility is available to anyone willing to fill out a form.

9. Disambiguation protects you. There are other people with your name. One of them is a minor-league pitcher and one of them is in the news for reasons you would not enjoy. Systems separate identical names using surrounding detail: city, employer, credentials, field. Every consistent listing adds another distinguishing feature and lowers the odds that a machine merges you with a stranger. A study of AI-generated answers found a meaningful share of confident attributions pointing at the wrong entity entirely; that is a coin flip you can load in your favor.

10. It compounds. One listing is noise. Twelve consistent listings across independent sources are a pattern, and pattern is what these systems reward – they are, at bottom, elaborate machines for noticing that many places agree. This also explains why the shortcut fails: dumping your name into fifty junk link farms produces volume without independence, and low-quality sources get discounted or filtered. Ten places that a human would respect beat a hundred that nobody would.

None of this is a growth hack. It is the same advice a decent mentor gave a nervous junior colleague in 1985 – join the association, go to the conference, get your name in the program – with one new wrinkle. The room now includes a very literal-minded participant who reads every roster, remembers every agenda, and forms opinions about who counts based entirely on where your name shows up and how often the sources agree.

So audit yourself the way a machine would. Search your own name and your company together and see what a stranger would conclude. Fix the listings that disagree. Renew the memberships you let lapse. Take the panel slot, even the 8:15 a.m. one on the last day.

Juliet was wrong about names, but she was right about one thing: it is a terrible idea to depend on a single source of information. Go get listed.

There’s a moment in Mad Men where the ad man tells a panicking client that when the conversation is going badly, you don’t argue with it. You start a different one. That was about cigarettes in 1960. It’s a pretty good job description for your website right now.

Here’s what nobody warned you about with AI search. When a buyer asks an assistant whether they should go with you or the shop across town, the assistant does not visit your homepage and admire your mission statement. It goes looking for text that already frames that exact choice. The head-to-head. The “which one is better.” The ranked list. And it finds one. It always finds one. The only real question is who wrote it.

For most companies, the answer is: a forum thread from four years ago written by somebody who had one bad week with you. A review aggregator whose business model is selling you a rebuttal. A directory that charges for placement and doesn’t mention that. Or, best case, a competitor’s marketing team who was clever enough to write the comparison first and generous enough to include you in it, in the column where you lose.

None of this advice is new in spirit. Researchers looking at how these systems pick what to cite keep landing in the same place: the machine surfaces what’s clear, current, specific, and easy to lift out of the page in one piece. Marketers have been saying a version of this since the first assistant started eating clicks. What is new is that a real chunk of your buyers now never land on your site at all, so your copy has to survive being read by software, compressed into three sentences, and repeated to someone who will never check.

So write the comparison yourself. Ten ways to do it.

1. One page per rival. Not one page listing eight of them. A page that lists eight competitors is a page about nothing, and a retrieval system treats it that way. Give each real competitor its own URL, its own headline with both names in it, and 600 to 900 words of actual comparison. Six of these beat one big roundup every time, because each one matches a question somebody is actually typing.

2. Lead with the question. Your headline should be the sentence a buyer would say out loud, not a clever phrase your agency likes. If people ask which is cheaper for a small team, that’s your H2, word for word. Then answer it in the first two sentences underneath, before any setup. Front-load the conclusion. Machines pull the paragraph directly under the heading, and so do skimming humans.

3. Concede something real. Say plainly where the other guy is better. Not a fake weakness, a real one. This does two things: it makes the rest of the page believable to a human, and it gives the model a balanced passage it’s far more willing to quote than a page that reads like a press release. Pages that only flatter themselves get treated as marketing and skipped. I’ve watched a client double their comparison traffic by adding one honest paragraph.

4. Best for whom. End every comparison with two short verdicts instead of one. Choose us if you’re this kind of buyer. Choose them if you’re that kind. Be specific enough that it stings a little: team size, budget, timeline, technical skill. Assistants love this structure because the question they’re usually answering isn’t “who is best” but “who is best for me,” and you’ve just handed them the answer key.

5. One yardstick, both ways. Pick five or six criteria that buyers actually raise and run both companies through all of them, in the same order, in prose. Not a chart. Charts get stripped out when a page is parsed, and half the meaning goes with them. Write it as short subheads with a paragraph under each. If you cover response time for yourself, cover it for them too. Uneven coverage reads as dodging, to people and to software.

6. Put a number on price. Everyone wants to write “contact us for pricing” and everyone gets punished for it. You don’t have to publish a rate card. Publish a range, a typical project, a starting point, what drives it up. A page with a real number in it becomes the source; a page without one becomes the thing a Reddit guess gets cited over. If your pricing is genuinely complicated, say what makes it complicated. That’s an answer too.

7. An alternatives page. Separate from the head-to-heads, build one page titled around the phrase people search when they’re leaving somebody else in your category. List the real options, describe each fairly in a short paragraph, include yourself, don’t rank yourself first. It feels insane the first time. It works because it’s the exact shape of the question, and because a page that treats the reader like an adult gets cited more than one that treats them like a lead.

8. Make switching concrete. Nobody stays with a vendor they like. They stay because leaving looks like a nightmare. So describe the leaving: what gets moved, who does it, how long it takes, what breaks, what it costs. A page that walks through a switch step by step answers a question no competitor is answering, which means it has almost no competition in the index.

9. Date everything. Put a visible “last reviewed” line at the top, and actually review it on that cadence. Stale comparisons are worse than no comparisons, because now you’re the one publishing wrong information about somebody else. Retrieval systems weigh freshness heavily on anything that looks like a product claim. A quarterly calendar reminder is the whole implementation.

10. Mine your sales calls. Stop guessing what the comparison should cover. Go pull the last thirty deals you lost and the objections you heard, and write the pages in the buyer’s vocabulary instead of your category’s vocabulary. They don’t say “end-to-end solution.” They say “I don’t want to babysit it.” Use their words. That phrasing match is most of why one page gets surfaced and a better-written one doesn’t.

Two things about posting it. Give each page a clean, boring URL with both names in it, keep the whole comparison on one page rather than behind tabs or accordions that hide the text, and add basic FAQ markup so the questions and answers are machine-readable. And don’t let these pages sit in a dusty resources folder. Link them from your pricing page and your navigation, mention them in sales emails, and say the same things in the places where your buyers actually argue, so the claim shows up in more than one house.

One caution, since somebody always takes this too far. Comparative copy is legally fine when it’s accurate and it’s opinion where it’s opinion. It stops being fine the second you invent a fact about somebody else’s product. Write it the way you’d want a competitor to write about you: firm, specific, a little generous, and true.

The alternative is leaving the sentence about you to a stranger, a scraper, and a directory getting paid by the person who wants your customers. That conversation is already happening. Go change it.