Rebecca Saltzburg

My strategy for custom-building AI SEO project skills

I custom-build an AI-driven skill for each SEO task, so a whole team produces senior-level work every time.

The problemThe senior decisions and the guardrails that catch the expensive mistakes, lived in people’s heads. Quality depended on who did the work.

• Each skill takes real exports and applies guardrails that stop bad calls automatically. I write the rules, the automation runs them and I review every judgment call.
• QA is built in. A compliance pass and a client-perspective pass, role-playing a marketing VP with no in-house SEO, run before anything reaches a client.
• Twelve-plus deliverable types are standardized onto one specification, with a data-integrity rule that reconciles every number to its source.

What it producesReusable skills plus the compliance layer the whole team runs on.

Sitewide content audit

A content audit that doubled an 800-page site’s organic traffic. It scores every URL as prune, update, or keep, with the reasoning attached.

The problemHundreds of URLs with no systematic way to decide prune, refresh, or keep and redirect targets picked by hand.

• A pipeline resolves crawl data, backlink data and Search Console into one row per URL, with canonicals and redirects resolved and the crawl as source of truth.
• A composite score built from seven weighted signals sorts every URL into prune, update, or healthy, each with a concrete redirect target.
• Five guardrails override the score outright. A page carrying real inbound links is never pruned on a low score alone.

What it producesA scored spreadsheet a non-SEO team member can act on. Every row names the target keywords, the redirect destination and the reason.

AI search: GEO & schema

Early adoption of AI search and GEO. The trust signals and structured data that get a page cited in AI answers and pulled into featured results.

The problemOn pages where trust matters, in health, finance and technology, the site often does not show who is behind the content and the code does not tell a search engine what the page is. It ranks lower and AI tools skip it.

• The E-E-A-T pass rates whether a page shows a real author, credentials, review and sources and lists the exact fixes.
• The schema pass reviews structured data by page type and hands back markup the team pastes straight in.
• That code is what gets a page picked up for featured snippets, voice search and AI answers.

What it producesA findings sheet plus ready-to-paste markup, organized by page type.

SEO content strategy, end to end

I own the whole content function. I ran 100% of PlushCare’s content, from keyword research to published, ranking pages.

The problemContent gets split across freelancers and tools with no single owner, so it is inconsistent, off-strategy and rarely ranks or converts.

• Keyword research, planning and the editorial calendar: every page mapped to real search demand before a word is written.
• Content creation with built-in E-E-A-T: expert and doctor review, credible citations and the trust signals YMYL pages need.
• On-page SEO and internal linking built into each piece, not bolted on afterward.
• Repeatable systems and templates so the whole operation scales without losing quality.
• Identifying and optimizing high-converting landing pages, so traffic turns into leads and revenue rather than just visits.

What it producesA full content program a brand can run on: strategy, calendar, briefs, published pages and the systems behind them.

SEO content outlines and keyword strategy

A keyword-mapped outline process that took a single page to 40+ top-three rankings and roughly $7,000 a month in traffic value.

The problemWriters got a keyword list with no structure, so a page chased one head term, missed the cluster around it and ranked for little.

• Each outline maps longtail, midtail and high-volume keywords into one structure, so a single page captures the whole cluster instead of one term.
• It builds in the AI entities and topical coverage that search engines and AI answer engines now expect.
• The writer gets the exact headings, questions and entities to cover, so the draft ranks without a second pass.
• Each outline builds in credibility and E-E-A-T: footnotes, expert or doctor review, citations to high-authority publications and studies, plus data-driven graphs, visuals and before-and-after comparisons.

What it producesA ready-to-write outline a writer executes directly, structured to rank for the full keyword cluster rather than a single term.

On-page optimization, rebuilt with AI

On-page rebuilt from an outdated tool into a custom, AI-driven process for each page.

The problemThe process ran on outdated software that produced the same generic boilerplate for every page.

• I had the team scrap the tool and move to AI-driven competitor research plus semantic topical authority analysis, combined with Search Console and backlink data, built live per page rather than from a template.
• Pages already ranking in the top three get flagged protect, so recommendations never dilute what already works.
• Every recommendation maps to an exact heading with a volume and cluster weighting: new H2, H3 or FAQ, or skip.

What it producesAn analysis document an editor executes without follow-up questions.

Link building and PR outreach

A rebuilt outreach process that improved link outreach response rates 200–300%. It defines where to pitch and the angle that earns the link.

The problemOutreach got a generic list of sites and guessed at angles, so links either could not be placed or landed somewhere that passed no value.

• I derive eight to ten outreach verticals from the client’s real offering, read off their navigation and sitemap, ordered broadest first, cutting verticals that are saturated or have no reachable sites.
• Every angle has to have a nameable link moment: a data, destination, definition, or example hook matched to the target page type.
• A reachability pass confirms each example site actually places a live, followed, in-content link to an author’s own domain. Two approval gates and a red-team pass verify every causal claim before anything ships.

What it producesA sourcing document the outreach team pitches straight from.

Data hooks and link context

The editorial reason a backlink belongs on the page.

The problemA pitch can name a great site and still give the editor no reason the link belongs on that page. So it never places, or it places without value.

• I match the hook to the page type, so linking to the client is the natural thing for an editor to do.
• Data-hook angles get built from an asset or a figure only the client has, something genuinely citable rather than repackaged.
• A red-team step verifies the causal claim the angle asserts and whose statistic it actually is, before outreach pitches it.

What it producesThe angle and hook engine built into the sourcing document.

Internal link audits

A standard for auditing in-content links and how they are counted.

The problemIn-content anchors were weak or generic and link counts were inflated because navigation, footer and menu links were being counted as links.

• A crawl export, regex filtered so only in-content editorial links count. Navigation, footer, CTA and menu links are stripped out before anything is counted.
• Generic anchors and branded-only anchors get flagged, with primary and secondary anchor language recommended in their place rather than raw counts.
• Every source page is verified live and present in the sitemap before it is recommended, with full URLs rather than slugs.

What it producesA three-tab spreadsheet: anchor fixes, redirect inlinks and links to add.

Anchor text strategy

A rebuilt anchor process that runs clean across every client with no setup.

The problemWriters over-use exact-match anchors, the clearest artificial-link signal there is. One or two primary keywords repeated across every placement builds a pattern a search engine can see.

• The method keeps the approved keyword, adds one neutral word and varies the grammar. No invented head terms. My own first version harvested vocabulary off the page and produced plausible-but-wrong anchors, so I rebuilt it.
• Guardrails derive automatically from the target URL, so one skill runs across every account with no per-client setup. It caps exact-match, bars generic anchors and respects retired anchors by reading strikethrough formatting in the source sheet.
• Medical targets carry extra rules: no outcome promises, no superlatives, exact clinical terms.

What it producesA paste block plus a rotation pool that feeds the anchor automation. Errors went from eight and seven in manual rounds to zero at full scale.

Cannibalization audits

A method for deciding which page wins and never redirecting one that still earns traffic.

The problemTwo or more pages competed for the same query and redirects were being made that killed pages still earning real traffic.

• The keeper is picked by search volume, existing traffic, link equity and content quality, not by guesswork.
• Title and H1 overlap gets mapped across the competing pages to show the actual collision.
• Every redirect candidate goes through a mandatory Search Console check before it ships.

ResultThe check caught a page earning 113 clicks that was about to be redirected away. I built the verification gate that catches it now.

What it producesA keeper decision and a verified redirect map.