Buyers stopped searching. They started asking.

You spent years building.

Buyers spend three seconds asking.

The window where someone typed a query into Google, scrolled, clicked through to a comparison post, and then to your homepage, that window is closing. It is not closing slowly. The replacement is already in pockets and on laptops, and the replacement does not click through. It answers.

When a buyer asks ChatGPT what is the best invoicing tool for a solo consultant who needs to chase clients in three currencies, the answer is a paragraph. Three names. Maybe a sentence on each. There is no second page. There is no scroll. If your product is one of the three names, you exist for that buyer. If it is not, you do not.

This is the work now.

The old playbook still ships. It just stopped landing.

You can rank first on Google for best CRM for small agencies and still lose the deal. The buyer never typed that into Google. They typed it into Claude. Claude gave them four answers. None of them were you. The traffic report on your SEO dashboard looks fine. The pipeline does not.

The mistake is reading the dashboard and concluding the work is fine. The dashboard measures the old surface. The new surface does not show up in it.

What changed under the hood

LLMs do not browse your site at decision time. They were trained, fine-tuned, and grounded on a corpus, then layered with retrieval that pulls from a different and smaller set of sources at inference. When a model writes a paragraph about your category, it is composing from what it saw during training plus what it can pull live from a handful of trusted destinations.

You can be in either set. Most products are in neither.

Training data is a slow lever. Retrieval is a faster one. The retrieval layer favors:

1. Sources the model trusts. High domain rating publications, named directories, Wikipedia, Reddit threads with engagement, structured comparison pages.
2. Content the model can parse cleanly. Headings that match the question, prose that names competitors by name, schemas that disambiguate your product from a vitamin brand with the same word in it.
3. Recency signals that say "this is current." A 2022 comparison loses to a 2026 one even if 2022 was more thorough.

You do not control the model. You control what the model can find about you when it goes looking.

DR is terrain. Not a strategy.

Domain rating still matters. It matters less than it did, and it matters differently. The model treats a high-DR source the way a human treats a quote from a known publication. The quote gets in. The unknown blog does not.

This does not mean chase DR for its own sake. It means: when a citation lands on a domain the model already trusts, the citation carries weight. When it lands on a fresh subdomain with no inbound links, it gets discounted or ignored. The same content has different value depending on where it lives.

The shift in tactic: stop optimizing your own pages for Google ranking. Start placing concrete, useful, named content on the sources the model is already pulling from.

The five seconds that matter

A buyer asking an LLM gets one paragraph. You have five seconds of attention inside that paragraph to do one of three things:

• Be named.
• Be named correctly.
• Be named for the right job.

Being named correctly means the model spelled it right, linked it right, and described what you do without confusing you with another product. Being named for the right job means the model surfaced you in answer to a question your product actually solves, not one it does not.

Both fail more often than they succeed. The fix is not louder marketing. The fix is denser signal: more pages, in more trusted places, that say the same true thing about your product in slightly different words.

What to do this week

Three concrete moves, in order of impact:

1. Scan what the models say about you today. Ask ChatGPT, Perplexity, Gemini, and Claude your top five buyer questions. Read the answers. Note which ones name you, which ones name a competitor in your place, and which ones get the description wrong. That gap is the work.
2. Find the sources cited. Most LLM answers include links or named publications. Those are the destinations the model is pulling from. If you are not on them, that is the next placement.
3. Write the comparison the buyer is asking for. Not a generic Findable vs Competitor page. The specific question. Best invoicing tool for a solo consultant chasing payment in three currencies. The page that answers that exact question, hosted somewhere the model already trusts, is the one the model cites.

This is not new content marketing. It is the same craft pointed at a different audience. The audience is the retriever.

Where this goes

The buyers are not coming back to search. The dashboard you have been reading is measuring a smaller and smaller fraction of the question. The new dashboard, the one that measures whether you exist on the surface buyers are using, does not exist yet on most products' walls.

Findable is the dashboard. Scan first. Fix what the scan finds. Publish so the model has something to cite.

Findable.