Tuesday, October 27, 2026

How to Lose at AI Search

Eighteen months ago I was a big fish in a small pond. AI search was a curiosity, the pond held a handful of practitioners, and a man who published daily, in his own name, on his own site, could be the answer the machines gave. Then everybody invested. The pond is an ocean now, the fish are innumerable, and I am small in it. I tell you this because the tactics changed when the water did, and most of the advice still circulating was written for the small pond. Here is the avoid list I work by now, in three tiers of worsening trouble.

Doodle in red, green, and blue ink: a large fish alone in a small pond, beside a small fish lost in a school inside a far larger pond

Tier one: the played out

These cost time and return nothing. Writing to a word count. Tuning keyword density like a radio dial. Publishing the same listicle ten thousand other sites published the same week, on the same topic, from the same outline the models hand everyone. Commodity content trades like pork bellies and frozen orange juice: interchangeable, priced by the ton, forgotten by Friday. The machines ingest it and cite none of it, because a claim stated identically in ten thousand places belongs to no one and differentiates nothing. If your page could be swapped with a competitor's page and no reader would notice, you have written feedstock.

Tier two: the folklore

These circulate at conferences and in group chats, repeated with confidence, supported by anecdotes. An llms.txt file as a ranking charm. FAQ schema as a resurrection spell for traffic the Helpful Content era took. Quoting yourself, stuffing statistics into every third sentence, and calling the result GEO. Buying citations in roundups assembled for that purpose. Each of these seems like it should work; each has the shape of a mechanism. The test is simpler than the folklore: the answer engines cite sources they already trust, and trust accrues to sites with a record. A charm file is not a record.

Tier three: what will kill you

Modern black hat wears better clothes than the old version and dies the same death. Content farms at machine scale, a thousand pages a day of fluent nothing, attached to a domain with a reputation the farm did not earn. Parasite hosting: your authority rented to a coupon page, a casino page, a pill page. Expired domains bought for their backlink profiles and refilled with a new business wearing the old one's skin. Reviews written by the reviewed. Pages that show the AI crawlers one text and human visitors another, with instructions to the machines hidden in the markup like a note passed in class. Every one of these works for a season. Then the update arrives, the site is gone from the index and the answers both, and the season's revenue is revealed as a loan.

The one asymmetry worth money

The chatbots answer from memory, and their memory runs months behind the world. Google's AI answers from the live index; it read this morning's pages this morning. A site that publishes current, accurate, first-hand material competes in the one arena where freshness is a ranking input rather than a courtesy. That is the whole business model of this weblog: show up daily, say something true that was not true yesterday, and let the up-to-the-minute machine find it. Stale authority defends itself in the chatbots. It starves in AI Mode.

So the positive program fits in one breath, and it is the close of every tier above. Earn the E-E-A-T signals the slow way: named expertise, first-hand experience, work other people cite because it helped them. Publish content that is useful, current, and different enough to belong to you. Avoid the played out, ignore the folklore, and never touch the third tier. The pond is big now. The fish who last are the ones the water trusts.

Two companion guides carry the same list in other uniforms: the owner version, written for the business deciding what to refuse, and the practitioner taxonomy, which names the modern schemes after the antiques they descend from.

Update, October 27, 2026, 9:35 AM: The prompt box now carries Chris’ dictation verbatim, per house rule, and E-E-A-T in the close now links to the glossary, which carries the term.

The prompt for “How to Lose at AI Search”

Breakfast is in I want you to make three blog posts two guides one on Christopher abraham.com one on g e r r i s c o r p.com and one on Chris abraham.net and they will all be various spun different versions of the same thing which is instead of focusing on what you should do in order to optimize for SEO and AEO and Geo make it a blog post on what to avoid and make it a three-tier list like what you should avoid what you shouldn't do and what will kill you kind of thing you can completely have three completely different unique takes on that but I want it to be you know things like what is the modern version of black cat what is the modern version of short game what is the modern version of played out what is the modern version of rumors and gossip that actually doesn't work what is the version of seems like it would work but doesn't work and and then if you want to have a final closing paragraph where you tell people that eeat is probably the best way to do it and that content forward and usable and different and unique and also current especially for Google's AI tool don't forget all the other models are 6 months behind in terms of their modern news like their their baseline is is in the past but modern search engines like Google their AIS are up to the minute and if you want to aim for still winning on Google because there hasn't been that much acceptance of AI search then you really need to make sure that well I don't know if this is true but if keeping up to date with the latest news on your website like I'm like we're doing on Chris abraham.net maybe that's usable because once you end up going from premium information to what's the term for pork bellies and frozen orange juice commodities you you start swimming in it and it completely different lake of content and this is exemplared from the fact that people like me who were early adopters of AI as search kit katnip now realize that when they when we started 18 months ago we were a big a big fish in a small pond and now that everybody has invested in it we are a small fish in a big pond can you make three completely different interpretations of the same of the same request? So that they don't cannibalize?