Technical SEO and GEO Signals Most Teams Miss
Eleven technical SEO and GEO signals that break quietly on B2B SaaS sites, from robots.txt to hreflang, schema and dates, each with a check and a fix.
Short answer: Technical SEO and GEO (generative engine optimization) stand on the same technical ground. A crawler has to be able to read your page, work out which URL is the real one and trust the information on it. Google says there are no additional technical requirements for AI Overviews and AI Mode: a page needs to be indexed and eligible to be shown with a snippet. So the technical side of AI visibility is mostly the signals you already know, set up correctly. The real problem is that these signals break quietly. No error appears; your page simply stays out of the answers. This post walks through the eleven signals B2B SaaS sites skip most often, each with a concrete check and a fix.
A SaaS website is rarely one thing. The marketing site runs on a JavaScript framework, the docs live on a separate subdomain and the blog sits in another CMS. Each part belongs to a different team, and the signals drift at those borders. We hit several of the items below on our own site, and we say so where it happened.
This post is about infrastructure, not measurement. We covered how AI visibility is tracked in our AI visibility monitoring guide and how visibility tracking differs from engine to engine in our post on engine differences.
1. Content exists only after JavaScript runs
What it is: If the text on a page appears only when JavaScript runs in the browser, the raw HTML the server sends is close to empty.
Google's JavaScript SEO documentation says Google processes these pages in three phases: crawling, rendering and indexing. A page that needs JavaScript goes into a separate render queue. It may stay there for a few seconds, and it can take longer. The same page says server-side rendering or pre-rendering is still a great idea, because not all bots can run JavaScript.
How to check: Pick a sentence from the first paragraph and search for it in the raw HTML: curl -s https://yoursite.com/page | grep "your sentence". If nothing comes back, a crawler that does not run JavaScript cannot see that text.
Common mistake: Looking only at Google's URL Inspection tool and concluding "the page renders". Even when Google renders it, other bots that do not run JavaScript still miss the text.
Fix: Server-side rendering, static generation or pre-rendering. Our site is built with React. During the build we produce static HTML for every route. Most pages are opened in a headless browser and saved; blog posts are turned into HTML directly by a separate script. Bots that do not run JavaScript still get the full text.
2. robots.txt group matching
What it is: robots.txt rules belong to user-agent groups. The standard (RFC 9309) says a crawler uses the group that matches its own name. It falls back to the * group only when no group matches it. When Allow and Disallow conflict, the longest, most specific rule wins.
How to check: Does your file have a separate User-agent group for any bot? If so, are the Disallow lines from the * group repeated in that group too?
Common mistake: We made this one. We used to have a separate "Allow: /" group for each AI bot, while the Disallow lines that closed directories such as /api/ sat only in the * group. Bots with their own group never read those rules.
Fix: We merged the bots into one group. The User-agent lines sit one under another, with the shared Allow and Disallow lines below them. The bot names stay in the file as an explicit permission signal, and the rules now apply to all of them.
3. AI crawlers do different jobs
What it is: "AI bot" is not one thing. OpenAI's crawler documentation says OAI-SearchBot is for search. A site that blocks it will not be shown in ChatGPT search answers, though it can still appear as a plain navigational link. GPTBot is for model training. The two settings are independent: you can allow OAI-SearchBot and disallow GPTBot. ChatGPT-User comes in on a user's request, and OpenAI writes that robots.txt rules may not apply to those visits. Search-related robots.txt changes can take about 24 hours to take effect.
On Google's side, Google-Extended is a product token. Google's documentation says it controls whether content may be used for training Gemini models and for grounding. It does not affect inclusion in Google Search and is not a ranking signal. It has no separate HTTP user agent.
How to check: Read your robots.txt bot by bot and write "search" or "training" next to each line. Check your server logs to confirm these bots actually visit.
Common mistake: Saying "we blocked the AI bots" and blocking the search bot along with them. Or assuming that blocking Google-Extended takes you out of AI Overviews. To limit what appears in AI features, Google points to snippet controls (signal 7); for its other systems, it points to Google-Extended.
Fix: Keep search bots open. Make the training-bot decision on purpose and write it down.
4. A robots.txt block hides noindex
What it is: Google's robots meta tag documentation says robots meta tags and the X-Robots-Tag header are only seen when a crawler fetches the page. If robots.txt blocks crawling, those rules are not read and are ignored. RFC 9309 adds two points. robots.txt is not a form of access authorization. And if the file returns a server error (5xx), the crawler must assume the whole site is disallowed.
How to check: List the URLs you have set to noindex and check whether any of them sit under a Disallow line in robots.txt. Then run curl -I https://yoursite.com/robots.txt and confirm it returns 200.
Common mistake: Closing an old pricing page or a staging environment with both Disallow and noindex. The crawler never sees the noindex. Google's canonicalization page also notes that URLs blocked by robots.txt can still be indexed without their content.
Fix: Leave the page you want out of the index crawlable and give it noindex. For content that truly has to stay private, rely on authentication, not robots.txt.
5. The canonical URL drifts
What it is: Google describes rel="canonical" and redirects as strong signals and sitemaps as a weak one. It advises against giving different canonicals for the same page through different methods, and recommends absolute rather than relative URLs. The JavaScript SEO page adds that the canonical is best set in HTML and that JavaScript should not change it to a different URL from the one in the original HTML.
How to check: Read the value in the raw HTML with curl -s https://yoursite.com/page | grep -i canonical. Then compare it with the rendered version in the URL Inspection tool. Both should match and point to the page's own URL.
Common mistake: A single-page app where every page's canonical points to the home page. Or English pages whose canonical points to the Turkish version. For a while our English pages were canonicalized to the Turkish ones.
Fix: Each language version should name its own URL as canonical. For localized pages, Google recommends specifying a canonical page in the same language. We switched our English pages to their own URLs.
6. hreflang pairs without return links
What it is: Google's documentation on localized versions says each language version must list itself as well as all other language versions. If two pages do not both point to each other, the tags are ignored. Alternate URLs must be fully qualified, including the protocol; a relative /en/page does not count. x-default is the page to show when no version matches the user's language.
How to check: Open the English URL listed in the Turkish page's hreflang set. Does that page's hreflang set include a line pointing back to the Turkish page? Are any of the URLs relative?
Common mistake: Printing hreflang on a single-language page that has no counterpart. Writing pairs with different slugs by hand and letting them drift over time.
Fix: We write hreflang during the build from one mapping. A page with no counterpart gets no hreflang at all. We moved to this after a site crawl found pairs without return links.
7. Snippet controls
What it is: Google's robots meta tag documentation says nosnippet applies to all forms of search results, including AI Overviews and AI Mode, and also stops the content from being used as a direct input for those features. max-snippet limits how much of the content may be used as a direct input for AI Overviews and AI Mode. Google's AI features page says a page has to be eligible for a snippet to be shown as a supporting link in those answers.
How to check: Run curl -s https://yoursite.com/page | grep -i 'name="robots"' and curl -sI https://yoursite.com/page | grep -i x-robots-tag. Has your CMS or a plugin added a site-wide nosnippet or a very low max-snippet?
Common mistake: Adding nosnippet to a pricing or security page "so nobody copies it". The page then cannot be a source in AI answers on exactly the topics buyers ask about most.
Fix: Apply restrictions on purpose and narrowly. Instead of closing a whole page, exclude only the part you need with data-nosnippet. Google documents this attribute for search snippets; we did not find its effect on AI features stated in the documentation.
8. Structured data and visible text diverge
What it is: Google's structured data guidelines say markup must be a true representation of the page content and that you should not mark up content that is not visible to readers. Markup that breaks the guidelines can lead to a manual action, which means the page loses eligibility for rich results. Even correct markup does not guarantee that anything will be shown. Google's AI features page also recommends making sure structured data matches the visible text on the page.
How to check: Start with the Rich Results Test and the URL Inspection tool. Then check by hand: do the FAQ answers, prices and dates in the JSON-LD appear on the page word for word?
Common mistake: FAQ schema with questions that are not on the page. A pricing page that was updated while the schema still carries the old price.
Fix: Generate the schema from the same data source as the visible text. On our blog posts, the FAQ is read from one field for both the page and the JSON-LD, so when the text changes, both change together.
9. The organization identity changes from page to page
What it is: Google's Organization documentation recommends placing this markup on the home page or on a single page that describes the organization. It does not need to be on every page. The url property helps Google identify the organization, and sameAs can list several external profiles.
How to check: Search every JSON-LD block on your site for "@id", "Organization" and sameAs. How many different @id values, name spellings and descriptions turn up?
Common mistake: We ran into this too. Our React pages shared one identity, while a hand-written static page used its own @id, a lowercase name and a separate description. The blog pages generated by a build script had a publisher node without an @id.
Fix: Pick one @id (for example https://yoursite.com/#organization). Other pages reference that identity instead of defining the organization again. The sameAs list comes from one place. The company name, address and phone number (NAP) read the same on the site, Google Business Profile, LinkedIn and directories. That last part is not a written Google requirement. It is our own consistency rule.
10. Dates do not reflect real changes
What it is: Google's sitemap documentation says it uses the lastmod value if it is consistently and verifiably accurate. The value should reflect the last significant update to the page. Changes to the main content, structured data or links generally count as significant; updating a copyright year does not. Google ignores priority and changefreq. Google's publication dates page asks that the visible date on the page match datePublished and dateModified.
How to check: Look at the lastmod values in your sitemap. If they all show the last build date, the signal is broken. Then compare the "Last updated" date on a post with dateModified in its JSON-LD.
Common mistake: Tying lastmod to the build date. Bumping the update date of every page because the footer changed.
Fix: For our blog posts, lastmod comes from the update field in the post's data file. Guides take their date from a data file too. Other pages first use the dateModified value in the page's structured data, and fall back to the last commit date of the page's own source file. The build date is never used. A change to a shared part such as the footer does not move that date. The date changes only when the page's own source file changes.
11. Treating llms.txt as a Google signal
What it is: llms.txt is a proposed text file that gives language models a summary of a site. Google's AI features page states plainly that you do not need to create new machine-readable files, AI text files or markup to appear in AI Overviews or AI Mode.
How to check: Does llms.txt sit above the first ten items on your technical to-do list?
Common mistake: Writing llms.txt and assuming the technical work is done. Leaving information in the file that the site does not say, or that is out of date.
Fix: We publish an llms.txt, but we do not treat it as a tool for Google visibility. Its one rule is that it says the same thing as the site. The priority is the ten items above.
Closing checklist
| Signal | Quick check | What good looks like |
|---|---|---|
| Content in raw HTML | Search for the first paragraph with curl |
Text arrives without JavaScript |
| robots.txt groups | Find bots with their own group | Disallow rules apply to every group |
| AI crawlers | Label each bot "search" or "training" | Search bots open, training decision written down |
| noindex vs Disallow | Compare noindex URLs with robots.txt | noindex pages are crawlable |
| robots.txt response | Check the status code with curl -I |
Returns 200 |
| Canonical | Compare raw and rendered HTML | Absolute URL, the page itself, same language |
| hreflang | Look for the pair's return line | Reciprocal, fully qualified, lists itself |
| Snippet controls | Read robots meta and X-Robots-Tag | No site-wide nosnippet |
| Structured data | Search the page for the schema text | FAQ, price and date match word for word |
| Organization identity | List every @id and sameAs value | One @id, one sameAs list, same NAP |
| Dates | Inspect lastmod and dateModified | Reflect real content changes |
| llms.txt | Compare the file with the site | Same facts as the site, not a priority |
Even with clean infrastructure, AI answers vary. We explain why in our post on answer variance and describe how we measure on our methodology page. If you would like us to run this audit and the fix cycle for you, see our GEO services page.
Limits of this post
Our sources are Google's and OpenAI's own documentation and the robots.txt standard. For Perplexity, Claude or Copilot, check those companies' documentation separately. These signals are a precondition for appearing in answers, not a guarantee. The documentation also changes: we quote it as we read it on November 13, 2026.
Sources
- [1] Google Search Central, "AI Features and Your Website"
- [2] Google Search Central, "Understand JavaScript SEO Basics"
- [3] RFC 9309, "Robots Exclusion Protocol"
- [4] OpenAI, "Overview of OpenAI Crawlers"
- [5] Google for Developers, "Google's common crawlers" (Google-Extended)
- [6] Google Search Central, "Robots Meta Tags Specifications"
- [7] Google Search Central, "How to Specify a Canonical with rel=canonical and Other Methods"
- [8] Google Search Central, "Localized Versions of your Pages"
- [9] Google Search Central, "General Structured Data Guidelines"
- [10] Google Search Central, "Organization Schema Markup"
- [11] Google Search Central, "Build and Submit a Sitemap"
- [12] Google Search Central, "Add a Byline Date to Google Search Results"
