Citation Engineer

Why it matters

AI answers get built from short, self-contained chunks (small passages the model pulls out on its own). A fact buried mid-paragraph, split across a table and a footnote, or written as marketing fluff is hard to lift and easy to get wrong. The clean unit changes by business. A distributor needs a spec stated plainly next to the part it describes. A roofing company needs a straight answer to 'how long does a roof last in a hailstorm zone.' A dental group needs 'does this office take Delta Dental' answered without a phone call. Same move every time. Put the question and the answer side by side, in plain words.

In the field

A dental group's 'new patient' page answered insurance questions inside a friendly welcome note. Asked 'does Bright Smile in Austin take Cigna,' ChatGPT said it wasn't sure. The fact was on the page, just wrapped in prose. Rewriting it as a plain line, 'We accept Cigna, Delta Dental, and Aetna at the Austin office,' got it quoted within weeks.

Edge cases

• A correct fact stated only inside an image or PDF the model can't read.
• An answer split between two pages, so no single chunk is complete.
• Marketing copy that talks around the fact instead of stating it ('industry-leading coverage' instead of the actual carriers).
• Two pages that state the same fact differently, so the model can't tell which to trust.

Why it matters

Plain words help. Labeled data removes the guessing. Schema markup tags a number as a price, a string as a brand, a block as a question and answer. It doesn't guarantee a citation, but it makes the fact machine-clear, which helps both classic rich results and AI retrieval. What you label differs by vertical. A distributor labels Product with the manufacturer part number and GTIN (the barcode-style global ID). A home-services site labels Service, service area, and reviews. A dental practice labels each office as a LocalBusiness with hours and accepted insurance.

In the field

A 12-location roofing site listed phone numbers and addresses as plain text. Adding LocalBusiness schema, one block per location with the real service area, made each office a distinct, machine-readable entity instead of twelve near-identical pages a bot blurred together.

Edge cases

• Schema that claims a rating or price the visible page doesn't show, which search engines treat as spam.
• Copy-pasted markup that leaves every location with the same address.
• Product schema missing the part number or GTIN, so the model can't match it to a real item.
• Markup added but the underlying fact is wrong, so you've just labeled an error clearly.

Why it matters

You can't be cited from a page no bot can read. Many sites block AI crawlers in robots.txt by default, hide content behind login walls, or load the key fact with JavaScript a bot won't run. This bites distributors hardest, where pricing and specs often sit behind a customer login, so the catalog is invisible to AI no matter how good the content is. Home-services and dental sites trip on the same wire when a chat widget or a booking script delays the real text.

In the field

A distributor's spec tables looked perfect in a browser but were drawn by JavaScript after load. PerplexityBot saw an empty shell. The fix wasn't more content. It was serving the spec table in the page's first HTML response so the crawler had something to read.

Edge cases

• robots.txt blocking GPTBot or ClaudeBot site-wide, often added by a security tool nobody remembers turning on.
• The fact behind a 'log in to see pricing' wall, so anonymous bots get nothing.
• Aggressive bot protection that returns a challenge page to crawlers instead of content.
• Content that loads only after a click or scroll the bot never performs.

Why it matters

On a big site you don't hand-edit pages. You fix the mold they're stamped from. Get the spec-block layout right in one product template and a quarter-million SKUs become quotable at once. The scale unit differs by business. A distributor has one template behind hundreds of thousands of product pages. A home-services company has a city-times-service grid. A dental group has an office-times-treatment grid. The trap is the same: one bad pattern in the template repeats the same citation failure everywhere.

In the field

A plumbing franchise had a page for every city-and-service combination, but each one buried the price range and response time inside a generic paragraph. Rewriting the single template so every page led with 'Drain cleaning in [city]: $X to $Y, same-day in most cases' made the whole grid answerable in one change.

Edge cases

• A template fix that looks right on the sample page but breaks on edge data (a product with no price, a city with no reviews).
• Manufacturer-supplied descriptions repeated across thousands of pages, so the model sees duplicate, low-trust text.
• Thin pages where the template has no real fact to structure, so you're polishing emptiness.
• A template change that fixes citation but quietly breaks the page's classic ranking.

Why it matters

AI grounds answers (checks claims against retrieved sources) and trusts a fact more when several independent sources agree. Your own page saying it isn't enough on a low-authority site. You need the same fact echoed where the model already looks: manufacturer sites, directories, review platforms, Wikipedia-class references, industry databases. The corroboration network (the web of outside sources that repeat your claim) is what turns a self-assertion into a citable fact. Where you build it differs. Distributors lean on manufacturer and interchange databases, home-services on local directories and review sites, dental on insurance directories and health listings.

In the field

A dental group claimed on its own site that it offered same-day crowns, but no AI mentioned it. The fix was getting that capability listed consistently across the practice's Google Business Profiles, two dental directories, and the device maker's 'find a provider' page. Once the claim appeared in several places, AI answers started naming the group for same-day crowns.

Edge cases

• Directory and profile listings that contradict each other (old address, wrong hours), which the model reads as low trust.
• Corroboration from sources the model doesn't weight, so effort produces no lift.
• A claim that's true but unverifiable anywhere outside your own site.
• Paid or spammy mentions that get the source discounted, taking your fact down with it.

Why it matters

Citations flow to the page that answers the question best, and the strongest answer is often a reference asset, not an essay. The data you sit on but never publish in usable form is the moat. A distributor owns part-number interchange data ('this Parker fitting equals that Gates one') locked in its PIM (the system that holds all product attributes). A roofing company owns real failure-mode and material-lifespan data from thousands of jobs. A dental group owns honest cost ranges by treatment. Publish it as a clean, structured table and you become the source the model cites.

In the field

A hydraulics distributor had thread-size and cross-reference data in its ERP that nobody outside the company could assemble. Published as a plain interchange table, one row per equivalence, it became the page Perplexity cited for 'what's the equivalent of [competitor part],' a query no marketing blog could answer.

Edge cases

• A genuinely useful dataset published as a PDF or an image, so no bot can extract a row.
• Reference data that goes stale (discontinued parts, changed prices) and starts feeding wrong answers.
• Aggregating someone else's data you don't have the rights to, instead of your own.
• A table so cluttered with marketing that the actual data is hard to lift.

Why it matters

At Senior the question shifts from 'fix this page' to 'which battles do we fight.' You pick the high-value queries (the ones a real buyer asks before a purchase). You track named mentions and share of voice (how often the brand is cited versus rivals) across ChatGPT, Perplexity, Gemini, and AI Overviews. Then you diagnose why a competitor gets cited when you don't. The program ties the Entry and Mid work to outcomes. Priorities differ by business. A distributor wants high-intent part queries, a home-services company wants 'best [service] in [city],' a dental group wants insurance and treatment-cost questions.

In the field

A regional roofing company tracked twenty buying-intent prompts across three AI engines and found it was named in two while a national competitor showed up in fifteen. The gap told them exactly which city-and-service facts to corroborate next, instead of guessing.

Edge cases

• Optimizing for prompts that get volume but no real buyers behind them.
• Treating one good answer as proof, when AI outputs vary run to run.
• Tracking mentions but never tracing why a competitor wins the citation.
• Chasing every engine equally instead of the one your buyers actually use.

Why it matters

Citation cuts both ways. An AI can name your brand and get it wrong: a wrong price, a discontinued product, a treatment you stopped offering, a pressure rating that's flat dangerous. You can't edit the model. You fix the inputs it grounds on. Correct your own page, update the directory or profile that's feeding the error, and add clear, current corroboration so the right fact outweighs the wrong one. This matters most where a wrong fact carries real risk, like a distributor's safety spec or a dental cost quote, but every brand needs the watch.

In the field

An industrial distributor found ChatGPT quoting a pressure rating for a hose that the manufacturer had revised down years earlier, a genuine safety problem. The old number lived in a cached spec sheet and two stale directory listings. Updating the page, the listings, and adding a dated correction got the answer to track the current, safe figure.

Edge cases

• A wrong fact sourced from a third-party page you don't control, so you must get that source changed.
• A correction that fixes one engine while the wrong answer persists in another.
• Stale data in an old cached copy that keeps resurfacing after you've fixed the live page.
• A safety- or money-sensitive error that needs fixing fast, before it spreads through more answers.