AI Search Specialist

Why it matters

There's no AI-search shortcut around the fundamentals. Senior GEO and AEO postings want years of prior SEO for a reason. Pages still have to rank, links still have to route authority to the right places, and a site that crawls badly is invisible to everyone, human or AI. The fundamentals don't change across verticals; only the page types do. A distributor lives or dies on category and product pages. A roofer lives on city and service pages. A dental group lives on office and treatment pages.

In the field

A six-office dental group rebuilt its site and quietly dropped the old URLs without redirects. Three offices fell off the first page of Google overnight. No amount of AI-search work would have helped. The rankings fire had to go out first.

Edge cases

• Treating AI search as a replacement for SEO instead of a layer on top of it.
• A redesign that breaks internal links or drops redirects and tanks existing rankings.
• Thin pages that were never going to rank, now blamed on "the algorithm."
• Chasing AI visibility while a core money page slides on Google.

Why it matters

Ranking and getting cited are different outcomes. Most studies find only a small slice of the URLs AI assistants cite also rank in Google's top 10 for the same question, somewhere around 12 to 15 percent. So a page can be quoted by an engine without ranking, and a top-ranked page can be ignored. Buyers reach you differently by vertical too. A procurement manager asks ChatGPT to compare suppliers. A homeowner asks Google's AI Overview whether a metal roof is worth it in Florida. A patient asks Perplexity which dentist nearby does same-day crowns. Same shift, three different question shapes.

In the field

An MRO distributor ranked #1 for its category but never showed up in ChatGPT answers about food-grade versus H1 lubricants. The engine kept citing a generic blog. Ranking wasn't the problem. Being citable was.

Edge cases

• Assuming a #1 Google ranking guarantees you show up in the AI answer.
• Confusing AI Overviews (Google's summary box) with a separate engine like ChatGPT.
• Reporting rankings as a stand-in for AI visibility when the two barely overlap.
• Ignoring that some engines answer from training data and others fetch live pages.

Why it matters

You can't improve AI visibility you've never measured, and the baseline is almost always worse than people think. The questions are vertical-specific, which is the point. A distributor audits part-number and substitution questions. A roofer audits "best roofer in [city]" and storm-damage questions. A dental group audits "[procedure] near me" and insurance questions. The audit doubles as the wake-up call that justifies the rest of the work.

In the field

A 40-location roofing company ran 15 "best roofer in [city]" prompts through ChatGPT and Perplexity. It got named in two. A regional competitor with a thinner website got named in eleven, because its reviews and citations lined up across the web. That gap became the roadmap.

Edge cases

• Testing made-up questions instead of the ones buyers actually type.
• Running the audit once instead of on a repeatable schedule.
• Counting a brand mention as a win when there's no citation or link.
• Forgetting that engines personalize and vary, so a single run is noisy.

Why it matters

You're one person, not two teams. Splitting the work is impossible, and it's also wrong. The same fix usually serves both goals. Clean structured data (machine-readable labels on your pages) helps Google understand you and helps an engine cite you. The leverage point differs by vertical. For a distributor it's the product template across 100,000 SKUs. For a roofer it's the city-page template. For a dental group it's the office-and-service template. Fix the template once and it lands everywhere.

In the field

A pump distributor was about to staff two roadmaps: an "SEO" backlog and an "AI" backlog. Both started with the same task: get spec data out of PDFs and into crawlable HTML tables. One roadmap, one fix, two wins.

Edge cases

• Running an SEO backlog and an AI backlog as competing queues that fight for the same hours.
• Optimizing a single page when the win lives at the template level.
• Doing the AI work while the underlying page still isn't crawlable.
• Letting "AI" become a separate budget line nobody can tie to an outcome.

Why it matters

AI earns its keep in production: drafting, cleaning up messy data, generating tables. The trap changes by business, but the failure is the same. A distributor's product text often comes straight from the manufacturer, so the identical paragraph sits on your page, the maker's page, and every competitor on the same feed. A multi-location roofer or dental group spins up city or office pages that are one find-and-replace apart. Engines see the duplicates and skip all of it. Deduping is the deliverable, not raw volume.

In the field

A dental group had 30 location pages that differed only by city name and phone number. Google indexed two and ignored the rest as duplicates. Rewriting each one around the real office, its dentists, its services, its parking, got them indexed and cited.

Edge cases

• Publishing AI-generated pages that are near-identical to each other or to the source feed.
• Treating word count as the goal instead of distinct, useful pages.
• Letting AI invent specs, prices, or claims that are wrong or non-compliant.
• Building facet or location pages before the underlying data is clean enough to tell them apart.

Why it matters

Every vertical sits on data engines want and can't find in a usable form. Distributors own interchange and cross-reference data, "Gates equivalent of Parker 387 hose," "1756-L61 replacement," buried in PDFs or a JavaScript lookup tool a crawler can't read. Home-services companies own real local answers: permit rules, material costs by climate, storm-season timelines. Dental groups own procedure, recovery, and insurance specifics. These questions have almost no reported search volume and almost total buyer intent. Publish the answer as plain HTML, not a PDF, not a widget, and you get cited where the bigger player can't even be read.

In the field

A small hydraulics distributor published its Parker-to-Gates interchange chart as a plain HTML table. Perplexity started citing it for "Gates equivalent of Parker 387," beating Parker's own cross-reference tool, which lives inside a JavaScript app the crawler never runs.

Edge cases

• Hiding your best data in a PDF or a JavaScript lookup tool the crawler can't read.
• Answering only the generic questions a thousand other pages already cover.
• Writing answers that need three paragraphs of setup before they make sense.
• Sitting on unique data in your PIM, ERP, or job records and never publishing it.

Why it matters

You can do everything right and still be invisible, for two reasons that cut across every vertical. First, content that only loads after a login or a script, like a distributor's contract pricing or a portal-gated service area, is content an engine never sees. Second, aggressive bot protection that blocks retrieval crawlers like OAI-SearchBot and PerplexityBot (the ones that fetch a page to support an answer) takes you out of those answers no matter how good the page is. This is senior work because it crosses marketing, IT, and security, and the fix is a template-and-config decision, not a per-page edit.

In the field

A distributor added strict bot protection at the edge and watched its AI mentions drop to zero. The security rule was 403-ing PerplexityBot along with the scrapers. One allowlist change, and the catalog was reachable again.

Edge cases

• Checking robots.txt but not the firewall or bot-management layer that actually blocks crawlers.
• Specs or prices that only render after a login or a script the crawler can't run.
• Blocking every bot to stop scrapers, and cutting off the retrieval crawlers too.
• Treating llms.txt (a proposed text file that lists your content) as if engines honor it. Google declined it, so use it as a complement to crawlable HTML, not a substitute.

Why it matters

Leadership sees the traffic chart falling and wants an answer they can act on. "AI Overviews" isn't one. The senior job is to separate what's truly lost from what got replaced by a citation, then brief it plainly instead of letting the chart speak. Because ranking and citation barely overlap, you can't use one as a stand-in for the other. You report both. This is also where the role becomes Head of Search, owning SEO, GEO, and AEO as one function and resourcing the split between them across whatever verticals the business runs.

In the field

A home-services company panicked at a 20% drop in organic clicks after AI Overviews launched on its core questions. The breakdown showed most lost clicks were research queries that now resolve in the answer box, while booking-intent traffic held and the brand's citation share inside the answers rose. Different story, calmer room.

Edge cases

• Reporting rankings alone and missing that you're absent from the AI answers entirely.
• Letting a falling traffic chart imply failure without separating lost from replaced.
• No fixed prompt set, so AI visibility is anecdote instead of a tracked number.
• Over-claiming attribution by tying a deal to a citation you can't actually trace.