Technical SEO, AEO and GEO Learning Roadmap for 2026

Three disciplines, one order of work

Technical SEO, answer engine optimization (AEO) and generative engine optimization (GEO) overlap more than their separate guides suggest. This roadmap gives a technical SEO specialist 12 weeks to learn all three on a real site, using free tools. The crawler is Seodisias, a desktop app for Windows, macOS and Linux that renders JavaScript, has no URL limit and needs no signup.

Each discipline answers a different question. Technical SEO asks whether a crawler can reach, render and index a URL. AEO, the practice of structuring a page so an answer engine can lift a clean, direct answer from it, asks whether that answer is easy to extract. GEO, the practice of making a page citable inside AI-generated answers, asks whether a generative engine will pick your page as a source. A page that fails the first question never gets tested on the other two, so the roadmap starts with crawling and ends with international and citation work.

Budget about six hours a week. Most of that time goes to reading crawl exports, not to running tools. Each fortnight ends with one artifact you can show a client or a manager: a PDF, a checklist or a spreadsheet.

Weeks 1-2: crawl health and robots.txt

Run Seodisias on a property of about 10k URLs and export the CSV. Record three numbers: how many URLs were crawled and rendered, how many are blocked by robots.txt, and how many crawled URLs are missing from the sitemap. Turn them into a one-page crawl health report.

Take a concrete case: a 10k URL store whose first crawl shows 1,400 faceted filter URLs, 300 of them linked from the main navigation. Those pages absorb crawler attention while product pages sit in the discovered-but-not-indexed bucket. Sort the export by internal link count to see which parameters get the most links, then decide per parameter whether to block, canonicalize or leave it open.

Robots.txt deserves its own hour. The format is defined in RFC 9309, which says a crawler follows the rules of the most specific matching user-agent group and treats an unreachable file differently from a missing one. A common mistake is blocking a path that the sitemap still lists, which sends Google contradictory instructions. Compare both files before you publish the report. If AI crawlers matter to your client, the robots.txt guide for AI bots covers the user-agent names. For a broader view of what a crawl export contains, read the complete SEO crawler guide.

Weeks 3-4: JavaScript rendering and Core Web Vitals

Crawl the same property twice in Seodisias, once with JavaScript rendering off and once on, and diff the two exports. Pages whose title, canonical or body text differs between the runs depend on client-side rendering, and those are your risk list. The JavaScript SEO rendering guide explains why each difference matters.

Then measure speed. Core Web Vitals consist of Largest Contentful Paint (LCP), Interaction to Next Paint (INP) and Cumulative Layout Shift (CLS). INP replaced First Input Delay (FID) as a Core Web Vital in March 2024, so a 2026 report that lists FID is out of date. Google’s web.dev Core Web Vitals documentation gives the thresholds for a good score, measured at the 75th percentile of real visits: LCP at or under 2.5 seconds, INP at or under 200 milliseconds and CLS at or under 0.1. Our own Core Web Vitals 2026 article walks through fixes for each metric.

Run each Lighthouse test three times and keep the median, because lab scores swing between runs. A typical finding is a hero image injected by JavaScript after the first paint: with rendering disabled the crawler sees an empty container, with rendering enabled it sees the image but LCP lands at 4 seconds. Note both numbers in the same table row so the cause is visible. Field data in the Search Console Core Web Vitals report lags by about four weeks, so judge your fixes on lab data first and confirm them in field data at the end of the program.

Weeks 5-6: structured data

Add JSON-LD for the types that match the visible content, using the vocabulary at schema.org. Start with ten pages. Pick pages where the visible text already answers a question in two or three sentences, since markup that describes content absent from the page fails validation or gets ignored.

Google removed most FAQ rich results from search, yet the markup still describes the page for other consumers, so the work is not wasted. A frequent error is copying one FAQ block across dozens of templates so every page claims the same questions. Give each page its own questions and answers. Validate each block with the Rich Results Test and log the result in a sheet with columns for URL, type, error count and warning count. The schema markup guide lists the types worth the effort.

Weeks 7-8: AEO and intent mapping

AEO starts with your own query data. Open the Search Console performance report, filter to queries phrased as questions and export them. Group the queries into informational, transactional and navigational intent, and write the label next to each URL in the crawl export so the mapping survives staff changes.

Keep the content calendar small. Four pages per fortnight is realistic for one specialist who also rewrites meta descriptions. For a query like “how to fix redirect chains”, the informational intent calls for a short definition near the top, a numbered fix list and one example with real status codes. A transactional query needs price, availability and a clear next step. Put the direct answer in the first two sentences under the heading, because that is the passage an answer engine can quote without editing.

Weeks 9-10: GEO and citation-ready pages

GEO is about being cited when a generative engine writes an answer. Test this by hand: ask the engines your clients care about ten questions from your week 7 list and record which sources they cite. Note what the cited pages share, such as a definition in the opening lines, a named author, dated figures or a table.

Then audit your own pages for the same traits. Seodisias includes an AI Ready check, and its AI Ready page lists what it tests. Pair that with the generative engine optimization guide to decide which fixes come first. If you use a language model to draft rewrites, treat its output as a draft: a model tends to open with the same verb across all five variants, so compare them side by side, keep each meta description near 155 characters, and fact-check every claim before it ships.

Weeks 11-12: international signals and local listings

Hreflang tells search engines which URL serves which language or region. Each annotation must point to a full, canonical, indexable URL of the alternate version, never to an internal identifier such as a content ID or a frontmatter field. Google’s localized versions documentation describes the three accepted placements: HTML link elements, HTTP headers and sitemap entries.

Hreflang errors tend to hide in return tags. If the English page points to the German page but the German page does not point back, search engines ignore the pair. Check that every version lists itself and every sibling, that x-default points at the English version, and that each page’s canonical points to itself. Use the crawl export to compare the lists side by side. Our hreflang implementation guide has the full checklist. For the Spanish version, use neutral Spanish so readers from Spain and Latin America both get a usable page.

If the site serves local customers, finish with the business listing. Listings are managed in Google Business Profile, and the Business Profile APIs are meant for managing existing locations at scale, not for creating a listing as a test. Name, address and phone must match character for character across listings, since “Street” on one and “St.” on another can create duplicate records. Test “near me” queries on a phone with location services on, from at least two spots in the service area, and write down the positions with the time of day. Local results shift within a few hundred meters, so a single screenshot proves little.

Measuring the program

Build the dashboard in Looker Studio, the current name of the tool formerly called Google Data Studio, and feed it with Search Console data plus the Seodisias CSV exports. Track four measures:

• Indexed share: indexed URLs divided by submitted sitemap URLs, from the Search Console page indexing report.
• Core Web Vitals: the share of URLs rated good for LCP, INP and CLS.
• Answer coverage: how many of your priority question pages have a direct answer in the opening lines.
• Citation checks: how many of your ten test questions cite your site, repeated monthly.

Set your own targets from the week 1 baseline instead of borrowing a figure from a blog post. Decide who owns each measure before the dashboard goes live: one person for indexed share, one for Core Web Vitals, and so on. Alerts that land in a shared inbox get ignored within a month. Store a snapshot of the dashboard at the end of every fourth week so you can show the trend from week 1 to week 12 without rebuilding the numbers from raw exports.

Common pitfalls

The first pitfall is treating crawling as the whole job. Teams that finish a crawl audit and stop there leave answer extraction and regional targeting untouched.

The second is crawling with a capped tool. A free tier that stops at 500 URLs shows you the first 500 pages in crawl order, which are the navigation and the oldest content, and it hides the deep faceted URLs where crawl budget is wasted. Screaming Frog’s free version has that 500 URL limit and reserves JavaScript rendering for the paid license, while Seodisias has no URL limit. A full crawl of a 10k URL property exposes those paths.

The third is shipping machine-written copy unchecked. Read every generated description aloud before it goes live.

The fourth is skipping regression checks. Rerun the same Seodisias crawl after each deployment and compare the exports line by line, so a single template change cannot undo weeks of work unnoticed.

Start the first crawl

Block two hours on your calendar, pick one property you can edit, and run the week 1 crawl with Seodisias. A roadmap run against a site you cannot change stalls in week 2.