Tag Archives: artificial intelligence

Déjà Vu, All Over Again: AI, Publishing, and the Delta House Rules

Leave a reply

Yogi Berra said it best. We’ve watched this exact movie twice – once in print, once in pixels.

1. The classifieds massacre was never really about Craigslist.

U.S. newspaper ad revenue peaked near $49.5 billion in 2005 and limped to about $9.8 billion by 2022 – an 80% haircut. Craigslist? Researchers put its damage at roughly $5 billion between 2000 and 2007. The bigger problem: publishers were marking classifieds up as much as 80% and treating it like an inheritance.

2. AI Overviews are the new Craigslist, only faster.

Seer Interactive found organic click-through rates on informational queries with AI Overviews fell 61% since mid-2024. Business Insider’s organic search traffic dropped 55% from April 2022 to April 2025, then it cut 21% of staff. Zero-click searches climbed from 56% to 69% in a single year, and Pew found 26% of people who hit an AI Overview ended the session there versus 16% who didn’t.

3. Everybody is on double secret probation.

Dean Wormer never told Delta the rules either. Today, opting out of AI Overviews means opting out of Google Search entirely. Which is less a choice than a ransom note. Otter said it cleanest: “You screwed up. You trusted us.”

4. We already mispriced the internet once.

Super Bowl XXXIV, January 2000: 17 dot-coms paid roughly $44 million for airtime. The following year, three showed up. The Nasdaq topped out at 5,048.62 on March 10, 2000, then fell 78% by October 2002 – about $5 trillion vaporized. The internet wasn’t the fad. The valuations were fat, drunk, and stupid.

5. The survivors owned the customer, not the channel.

Amazon fell roughly 95% and lived. Up to half of all dot-coms simply died. The dividing line was never who had the best technology – it was who owned the relationship when the traffic stopped showing up. Priceline lost about $30 a ticket and fell 99%, then rebuilt into a giant. The bust sorted business models, not industries.

6. A rented audience always gets evicted.

One hundred thousand newsletter subscribers beat a million drive-by clicks a platform can revoke on a Tuesday afternoon. U.S. newspapers have lost more than 2,500 titles since 2005 – most of them profitable right up until somebody else’s algorithm changed. Direct beats derivative. It always has.

7. “Was it over when the Germans bombed Pearl Harbor?”

It wasn’t over, and it also wasn’t accurate – and nobody stopped Bluto, because he was the only one in the room still moving. Publishing doesn’t need a better historian right now. It needs somebody willing to lead the parade with total conviction while the smart money sits on the curb.

The AI apocalypse isn’t coming for publishers who own their audience. It’s coming for the ones who rented one.

Publish the Post-Mortem: Marketing’s Empty Confessional

Leave a reply

Why the campaign that lost money is the one thing a competitor physically cannot steal.

Every Saturday afternoon, somebody sits in a wooden box and tells the truth about a bad week. No deck, no spin, no attribution model. Just what happened, how often, and what he intends to do about it. Marketing never built anything half that powerful, and the booth sits empty, door unlocked.

  1. The line is short. That is the whole opportunity.  Roughly 2% of American Catholics go to confession regularly; three-quarters never go or go less than once a year (CARA, Georgetown). Now count the firms in your industry publishing the campaign that lost money. Same arithmetic, weaker excuse.
  2. “We weren’t perfect” is not a confession.  Any priest sends you back out to try again. You need the sin, the number, and the frequency: “$18,400 on paid social across nine weeks, 61 leads at $302 apiece against a $75 target, shut off on day 63.” That detail is the difference between contrition and a press release.
  3. Confess the sin, not the weather.  Salancik and Meindl combed corporate annual reports and found firms that pinned bad results on outside forces posted weaker growth and profitability afterward. “Market headwinds” is corporate for “mistakes were made.” The algorithm did not betray you. You wrote the offer.
  4. A spotless record reads as a fake one.  Northwestern’s Spiegel Research Center found purchase likelihood peaks between 4.0 and 4.7 stars and slides as ratings near 5.0 – in no category was a perfect score optimal. Reevoo found shoppers who engage with negative reviews convert 67% more often and linger 5x longer, while 95% grow suspicious when nothing critical appears. Your all-wins case study page is a 5.0. Nobody believes a 5.0.
  5. “You too?” is the most persuasive phrase in business.  A 2026 survey of 1,500 U.S. Catholics found 43% of infrequent penitents would go more often if told that struggling with the same sin repeatedly is normal; 53% said awkwardness keeps them out. Your prospect torched a budget on your exact bad idea last spring and told no one. Be the first man to say it out loud.
  6. No penance, no absolution.  You do not leave the box holding regret. You leave with a penance and a firm purpose of amendment. So end the write-up with the rule you now follow because of what it cost. Skip that and you are not confessing, you are wallowing, and nobody hires a man still crying.
  7. You cannot confess another man’s sins – and there’s your moat.  A rival can lift your headline, pricing grid, and landing page before lunch. He cannot lift the Tuesday you watched $18,400 walk out the door. He was not in the room. Everything else you publish is one screenshot from being his. (Anonymize the client. The failure is yours to publish. Their name is not.)

Go on. The booth is open, the priest has heard worse, and the line is short.

When the Machine Lies About You – And Nobody’s Job It Is to Fix It

Leave a reply

You hired a spokesman you never interviewed. He works around the clock, sounds certain, and quotes 2025 prices.

It Isn’t Lying. It’s Repeating.

1. The damage is measurable. One audit found 72% of brands had at least one flat factual error in AI answers about them, usually stale pricing lifted from a review site that quit updating. The Faro Index scanned 244 companies across four assistants: 88.8% accuracy, one fact in nine wrong, fintech last at 82.4%.

2. Arguing with the model is arguing with Paulie Walnuts. Paulie originates nothing. He hears a thing at the Bing, repeats it with conviction, and by Tuesday it’s gospel. You don’t correct Paulie – you correct the room. NP Digital ran 600 prompts across six platforms: ChatGPT was fully correct about brands 59.7% of the time; Grok, 39.6%.

3. A citation is decoration, not proof. Only 51.5% of AI sentences were fully supported by their cited sources. The BBC tested 100 news stories: 51% of answers had significant problems, 13% of quotes altered or invented.

4. The playbook, five moves. (a) Build a claim ledger: every material fact next to its source of record. (b) Run a fixed prompt set monthly across the major assistants, same wording, logged. (c) Record the cited URLs; the URL is the crime scene. (d) Repair upstream: the dead directory listing, the pricing page nobody redirected, the PDF rotting on your server. (e) Retest in 30 days.

5. The leak starts at home. Livia never raised her voice. She just fed Junior a slightly wrong version of events, and people died over it. Most bad answers trace to something you published and never killed: an old deck, a zombie landing page, a partner site running 2019 boilerplate.

Nobody Owns This, So Name Your Consigliere

6. Right now it’s a no-show job. Tony was on the books at Barone Sanitation: title known, work optional. No job title exists for correcting what an assistant says about you, so it lands on whoever noticed. That is not an org chart.

7. Silvio had the real assignment. Know what’s being said, know if it’s true, get to it early. Name it – AI Answer Accuracy Lead – and give it to one person in marketing ops or comms. Committees notice things; people fix them.

8. The whole job, monthly, four to six hours. Run the prompt set. Update the ledger. Triage by severity times frequency (price beats founding year). Open source-repair tickets with an owner and a due date. Send one page upstairs: what changed, what’s still wrong, what it costs.

9. Skipping it is the Adriana problem. A quiet problem, known to one person, buried because nobody wanted the conversation. It compounded, and then came the meeting in the woods.

The Bottom Line

You can’t argue with the machine. You can change what it reads. Put a human name on that this month – until somebody owns it, it stays nobody’s job until it’s everybody’s emergency.

Resurface The Archive: Why Fred Flintstone Still Out-Earns You

Leave a reply

Fred hasn’t shot new material since April 1, 1966. He’s still on television. Your best piece is buried on page nine of your own site. Here’s the fix, and it takes an afternoon.

1. The people reading you now are not who you wrote it for

Audiences turn over constantly. Email lists rot about 2.1% a month – roughly 22.5% a year (MarketingSherpa, via HubSpot’s own decay simulator), and 30-35% in fast-moving sectors. Add normal follower growth and a two-year-old post faces an almost entirely new room. You’re not repeating yourself. You’re premiering.

2. Your best work had the shelf life of a delivered pizza

Median half-life of a tweet: 24 minutes. Facebook: 76 minutes. LinkedIn: 24 hours. YouTube: 6 days. A blog post: 2 years (Graffius, across 25+ sources). You spent nine hours on something that was structurally dead before your coffee cooled. That’s not a quality problem. It’s a distribution problem, and those can be re-run.

3. The Bedrock Precedent

166 episodes, nothing new since 1966 – and Fred has run in continuous syndication for sixty years. The kicker: the show was itself a resurfaced archive, The Honeymooners relocated to a rock quarry. Jackie Gleason considered suing and passed, not wanting to be known as “the man who took Fred Flintstone off the air.”

4. The data is lopsided, and nobody acts on it

HubSpot’s historical optimization lifted monthly organic search views on updated posts 106% and more than doubled their leads. Orbit Media, surveying 11,000+ bloggers, finds updaters 2.5x more likely to report strong results. Portent found full rewrites drove 454% keyword growth; cosmetic tweaks did nearly nothing. Depth decides the payoff.

5. The “here’s what I got wrong” paragraph is the whole play

Anyone can repost a win. Almost nobody publishes a receipt. A short, specific correction – the prediction that missed, the tool you recommended that got acquired and ruined – buys more credibility than three new articles of confident guessing. Fred spent six seasons certain; Wilma was right every time. The gap between Fred and a writer people trust is one honest paragraph.

6. Pick like a quarry foreman, not a sentimentalist

Three filters: highest historical traffic, oldest publish date, something you still believe. Resurface your heavy hitters, not your orphans – a post that flopped in 2023 flopped for a reason. Then: new headline, rebuilt opening, current numbers, dated “what’s changed” box up top.

7. Run the arithmetic before you write another cold open

The average blogger burns 3 hours 48 minutes on a new post, and per Ahrefs 96.55% of blog posts get zero Google traffic. That same afternoon, spent on a piece with existing links, rankings and proven demand, is betting on a horse that already finished the race. Fred would take that action. Fred would also lose the ticket.

Stop mining new rock. You’re standing on the quarry. Yabba dabba doo.

Come On Down: Your CRM Has No Idea Where Your Customers Came From

Leave a reply

Thirty interviews. Three weeks. A dataset nobody can copy.

Call thirty customers you already won. Ask where they first heard your name – not the form they filled out, the first time. Publish what they tell you. Here’s why it works, and why nobody does it.

1. The CRM records the podium. The real selection happened in the parking lot.

The Price Is Right sells randomness: lightning strikes, a stranger sprints down the aisle. Reality – 300 people fill each taping, nine reach the stage, and a producer already interviewed every one of them in line, sizing them up in five seconds. For four decades that was Stan Blits, hunting energy and humor. None of it airs. The broadcast begins at “come on down.” Your CRM is the broadcast.

2. The gap isn’t a rounding error.

Refine Labs published the comparison: their software credited web search for 79% of conversions. Customers credited search 3% – and put 98% of closed-won revenue on dark social: podcasts, communities, word of mouth. SparkToro found 100% of visits from Slack, Discord, and WhatsApp logged as “direct.” That’s not slightly off. That’s a different show.

3. By the time you get a touchpoint, the episode is already taped.

Gartner: B2B buyers spend 17% of purchase time with all vendors combined – 5–6% with any one rep. 6sense: 81% have a preferred vendor before first sales contact. Forrester: 92% start with someone in mind. That demo request isn’t discovery. It’s a formality.

4. Ask the question properly, then shut up.

“How did you hear about us?” on a form gets you “Google” – a hallway, not a room. Ask a human, after the close: “Walk me back. Where were you the first time you heard our name, and who said it?” Then stop talking. The answer is usually a person, a podcast, or a Slack channel you’ve never expensed.

5. Thirty is the number. Five is a story, a hundred never gets finished.

Twenty minutes each is ten hours – three weeks between other things. Below fifteen you’re guessing. Past forty you’re procrastinating with extra steps.

6. The Ted Slauson principle: the guy with the homework wins.

Slauson spent years taping episodes and memorizing prices – a dataset nobody else bothered to build. On the show taped September 22, 2008, Terry Kniess bid $23,743 on the Showcase. Exact. To the dollar. The first perfect bid in 36 years of daytime episodes. Producers cried foul and proved nothing, because the edge was never cheating. It was attendance.

7. Publish it – that’s the part competitors can’t touch.

A rival clones your landing page over lunch and outbids you on your brand terms by Friday. Thirty conversations with your customers, he cannot touch. Publish the delta: what the CRM claimed, what buyers said, the number that embarrassed you most. You get a defensible budget and an uncopyable post.

Stop guessing the actual retail price. Go ask the people who already paid it.

AI Insurance Is Evolving Exactly Like Cyber Insurance Evolved

Leave a reply

If you sat through a cyber renewal in 2021, you have seen this movie. It is Groundhog Day, the clock radio just went off, and the same guy is insisting the risk is unknowable. Only the editing is faster this time.

1. Silence Is Not a Coverage Grant. Cyber losses were once covered mainly because nobody had excluded them. The industry called it silent cyber. Now it says silent AI, the same coin flip, except the coin talks back.

2. Act Two Is Always the Exclusion. ISO’s generative AI endorsements, CG 40 47 and CG 40 48, took effect January 1, 2026, with a definition broad enough to swallow anything producing text, images, audio, video, or code. Cyber took a decade to get here. AI took twenty-four months.

3. Cyber’s Numbers Explain Cyber’s Manners. Loss ratios at major carriers crossed 100% in 2020 and 2021. Marsh clocked premiums up 79% in 2022. AM Best put U.S. premium up 50% to $7.2 billion, with the standalone loss ratio dropping 23 points to 43%. Carriers did not find religion. They found a spreadsheet.

4. The Application Form Became the Security Program. By 2021, MFA went from best practice to no MFA, no policy. Microsoft says it blocks over 99% of automated account-compromise attacks. A two-page questionnaire hardened more networks than a thousand keynotes. Answer it wrong and you get a decline, a sublimit, or a rescission at claim time.

5. AI Is Running the Same Play. Carriers now want a model inventory, human-in-the-loop review, a tested kill switch, a named accountable executive, and provable data provenance, mapped to NIST AI RMF, ISO/IEC 42001, and AIUC-1. AI sublimits near 10% of the limit are the polite way of saying we are unsure.

6. New Carriers Arrive Before the Actuaries Do. Armilla lifted its Lloyd’s capacity to $25 million in January 2026. AIUC came out of stealth with $50 million tied to its own audit standard. Testudo began underwriting in January 2026. Munich Re’s aiSure has guaranteed model performance since 2018. That is how cyber started.

7. Actuaries Have the Hardest Job in the Building. Nobody can price a tail event on twenty months of data, so the loss triangle is being built from court dockets: Bartz v. Anthropic, reportedly $1.5 billion over training data; Moffatt v. Air Canada, where a chatbot’s improvisation became a contract; Mobley v. Workday on algorithmic hiring.

8. What to Do Before the Next Renewal. About 74% of small businesses use AI; almost none have read their endorsement schedule. Is CG 40 47 or 40 48 attached to my policy? Do I have affirmative AI coverage, or just the absence of the word no? Can I prove my governance, or only describe it? “We told everybody to be careful” is a sentiment, not a control.

Murray escaped Punxsutawney only once he used what the repetition taught him. We already have the cyber tape.

When the AI Models Improve 10x: The one-sentence test most companies are about to fail in public

Leave a reply

One question now outweighs your entire content calendar: why does your company still exist once the AI models are ten times better and ten times cheaper? You get one sentence. No deck, no roadmap, no “we’re AI-native.” Most cannot. Saying yours out loud, with your name on it, is the cheapest advantage on the market.

1.The fastball keeps getting faster. Stanford’s AI Index clocked GPT-3.5-level inference falling from $20 per million tokens in November 2022 to $0.07 by October 2024 – 280x in 18 months. a16z calls it LLMflation: roughly 10x cheaper per year at equal quality. Moore’s Law doubled transistors every 24 months – a leisurely 1.4x a year. AI is lapping it.

2.Learn the other laws too. Wright’s Law: cost falls a fixed percentage each time production doubles – volume sets the price, not the calendar. Jevons Paradox: cheaper means more, not less. Amara’s Law: we overrate two years and underrate ten. And METR finds the length of tasks AI finishes unsupervised has doubled every 7 months for six years – nearer every 4 lately.

3.If your sentence starts with “we use AI to,” you don’t have a sentence. You have a feature, and features get absorbed like a September call-up. MIT’s 2025 review found roughly 95% of enterprise GenAI pilots produced no measurable P&L impact – not because the models were weak, but because the pilots were features in a trench coat.

4.Answer in public. It’s a moat and a magnet. Post what gets commoditized, what doesn’t, and why you’re still standing. Every competitor who dodges looks evasive by comparison. First and specific beats right and late.

5.As a company: own what nobody can download. Proprietary data, distribution, liability, the last mile, the relationship. A model will out-write you; it won’t sign your customer’s contract or take the blame at 2 a.m. Batting .300 puts you in Cooperstown – that’s failing 70% of the time.

6.As a website: be the source, not the summary. Summaries are free now. Ship original numbers, first-party data, named authors. If your page can be reassembled from three others, it will be – and nobody visits yours.

7.As a marketer: buy citations, not volume. The win isn’t a click – it’s being the name the model says. Move budget into primary research and a view worth quoting. Cheap content just became free content, and free content is worth precisely that.

THE PLAN (in batting order)

Write the sentence this week and publish it. Audit every product line against it – whatever breaks when the model gets 10x better, kill it or wrap it in something human. Shift 20% of content spend to original data. Measure citations, not clicks. Re-run it in 90 days; the pitch will be faster. Swing.