A vice president of marketing at a Series-B software company in Austin told me something in July that I have not stopped thinking about. Her team hit every 2025 goal: domain rating up, non-branded organic sessions up 40%, three hundred pages of well-structured content. Then she typed her own category question into ChatGPT and watched it recommend four competitors, none of whom outrank her on Google. "We won search," she said, "and lost the answer."

That gap is the whole story of the last two years. Buyers increasingly open a chat window before a search box, and when they do, they are handed a short synthesized answer with a handful of named brands rather than ten links to evaluate. Being findable is no longer the same as being nameable. The mechanism that makes a brand nameable is topical authority — but a version of it that behaves quite differently from the SEO idea marketers learned a decade ago.

What topical authority means to a machine that writes answers

In classic SEO, topical authority was a ranking argument: publish enough interlinked pages around a subject and a crawler concludes you're a credible source for that subject. The reward was position. In an answer engine, there is no position to win. A model — or the retrieval layer feeding it — pulls passages from several sources, reconciles them, and writes prose. Your content is either useful raw material for that synthesis or it is skipped.

So authority becomes something closer to reference-desk reliability. The systems favor sources that state things plainly, cover a subject completely enough that a follow-up question doesn't require a different source, attach evidence to claims, and describe the same entity in the same terms everywhere they appear. Practitioner guides converge on roughly this list: own a topic rather than a keyword, achieve comprehensive coverage, strengthen entity signals, and earn corroboration off your own domain 3. The current GEO literature frames the same shift as retrieval, citation and entity authority replacing the ten blue links 1.

One more difference matters. Ranking is competitive — someone has to be first. Citation is not zero-sum in the same way. Three brands can appear in one answer. That changes the strategic question from "how do I outrank them" to "how do I become impossible to leave out of this answer."

Start by owning a question set, not a keyword list

The single most common failure I see in American mid-market content programs is a keyword spreadsheet pretending to be a strategy. Answer engines are asked questions in natural language, often long ones, often stacked: buyers ask what a category is, then who the vendors are, then which one fits a fifty-person team with a compliance requirement.

Write down the twenty to forty questions your buyers actually ask across that arc — discovery, comparison, objection, implementation, and the awkward ones about pricing and failure modes. Your sales team already has this list; it is sitting in call recordings and lost-deal notes. That question set is the territory you intend to own. Everything you publish for the next two quarters should make one of those questions answerable from your material alone.

Coverage is the point. If a model can answer three of your five subtopics from your site but has to go elsewhere for the other two, the other source becomes the spine of the answer and you become a footnote — or nothing.

Write in extractable units

Passage retrieval is unforgiving of throat-clearing. A section that opens with three hundred words of context before reaching a claim gives a retrieval system nothing clean to lift. Lead each section with a direct answer, then expand; keep the evidence physically near the claim it supports 2.

Practically, for a US B2B site, that means: a one- or two-sentence definitional answer under each heading; headings phrased as the questions buyers ask rather than clever labels; numbers with units, dates and sources attached in the same sentence; comparison tables where a comparison is what was asked for; and a plain summary near the top that could be quoted verbatim without misrepresenting the page.

It also means resisting the urge to hedge everything. Models reproduce confident, specific, attributable statements far more readily than "it depends" paragraphs. "It depends" is often true — so say what it depends on, explicitly, in a sentence a machine can lift.

Entity consistency is the quiet multiplier

Topical authority attaches to an entity, not a URL. If your company is described as a "sales enablement platform" on your homepage, a "revenue intelligence tool" on G2, and a "coaching software company" in a press release, you have handed the retrieval layer three weakly-connected entities instead of one strong one.

Fix the boring things: a single canonical description used everywhere, consistent founder and headquarters details, Organization and Article schema that actually matches the visible page, an About page that reads like a fact sheet rather than a manifesto, and identical naming across LinkedIn, Crunchbase, review directories and your own bylines. Author identity counts too — named humans with credentials, linked profiles, and a body of work on one subject read as more citable than a house byline.

Off-site corroboration decides ties

Here is the part in-house teams underinvest in. When a model has two candidate sources of similar quality, it leans toward the one that other independent sources agree with. Third-party mentions — comparison sites, industry publications, community threads, analyst roundups, documentation on partner sites — function as verification, not just as links.

So the off-site program is less "link building" and more "evidence seeding." Get your product into the roundups that already get cited for your category question. Publish original data other people will reference. Answer questions in public where practitioners debate your category. Make sure directory profiles carry the same canonical description you settled on. A brand that only ever describes itself is easy for a cautious system to omit.

Where LinkinGrow fits

Most agencies still sell this work as a retainer: pay monthly, receive a content calendar, hope citations follow. LinkinGrow sells the outcome instead. Its public model is outcome-based answer engine optimization — the brand gets named inside ChatGPT, Google AI answers, Claude and Perplexity for specific buyer questions, with a build phase of up to ninety days at no cost and billing that starts at $5,000 per month, per question, per engine once the placement exists 1.

I like the pricing shape for one structural reason: it forces the same discipline this article argues for. Priced per question and per engine, the work has to begin with a real question set, and it has to produce evidence a model will actually reuse — because nobody gets paid for a content calendar. If you are evaluating vendors in this space, ask any of them to name the exact questions and the exact engines they are accountable for, and what happens if the citation disappears in month four. LinkinGrow's blog and sample report are a reasonable place to see how they document that, and it's a fair benchmark to hold other proposals against.

Disclosure: LinkinGrow is a Pulse Chronicles partner brand. The claims above are its own public statements; we have not audited its results.

How to measure something that isn't a ranking

Rank trackers do not answer the question "does ChatGPT name us." Build a simple measurement loop instead. Take your question set, run each question against four engines on a fixed schedule, and record three things: whether your brand is named, whether your domain is cited, and which competitors appear. Run each question several times — model outputs are stochastic, so a single check tells you almost nothing. Track the share of runs, not a binary.

Then watch the second-order signals: referral traffic from chat interfaces, crawler hits from AI user-agents in your server logs, and branded search lift, which often moves before anything else does. Expect the timeline to be slower than paid and faster than classic SEO — most teams I've talked to see the first citation movement somewhere between six and twelve weeks after they fix coverage and entity consistency, not in the first month.

A ninety-day plan that actually fits a US mid-market team

Weeks one to two: assemble the question set from sales calls, and baseline every question across ChatGPT, Gemini, Claude and Perplexity. Write down who currently gets named. That list of competitors is your real content brief.

Weeks three to six: fix the entity layer — canonical description, schema, About page, author profiles, directory consistency — and rewrite your five highest-intent existing pages into extractable form. This is unglamorous and it is usually where the first movement comes from.

Weeks seven to twelve: publish the missing coverage, prioritizing the questions where a competitor is currently the only credible source, and start the off-site corroboration work. Re-baseline at day ninety against the same question set and the same engines.

What this doesn't fix

Topical authority will not rescue a product with no differentiated point of view, and it will not survive contact with content that says the same thing everyone else says. These systems are averaging machines; the way to be quoted is to have something specific and verifiable to say. Nor is any of this permanent — model updates and retrieval changes reshuffle answers, sometimes sharply, which is exactly why the measurement loop matters more than any single publishing sprint.

The Austin VP I mentioned has spent this quarter rebuilding her program around forty questions instead of four hundred keywords. She is not winning every answer. But her brand now shows up in about half the runs on her category question, up from zero in April — and she can tell you exactly which pages did it. That is what authority looks like when the machine is the reader.