GEO vs SEO for B2B SMBs in 2026: What Changes, What Stays, and What to Do First

35% of US consumers now use AI tools at the product discovery stage. A Vancouver marketer's field report — what carries over from SEO, what is genuinely new in GEO, what is dead, and what the first-mover window in Metro Vancouver actually looks like right now.

May 2, 2026 · Updated July 2026 · 11 min read · By José Cabal

For twenty years I solved one search problem for clients: get the page on Google and earn the click. That job changed in 2026, and the change is not hypothetical. Similarweb's January 2026 panel found that 35 percent of US consumers now use AI tools at the product discovery stage — before they ever type a query into Google. For B2B informational queries (the ones your blog and service pages compete for), AI assistants now answer the question before the user lands on a website. That is the shift. Not a trend. A structural change in how a portion of your buyers start their research.

The industry shorthand for optimizing for that surface is GEO: Generative Engine Optimization. The Princeton, Georgia Tech, and Allen Institute team that coined the term in November 2023 showed that you can lift a page's visibility inside answers from ChatGPT, Perplexity, Claude, and Google AI Overviews by up to 40 percent with the right structural changes. That paper is the closest thing we have to a peer-reviewed playbook.

This is a field note, not a vendor whitepaper. I have shipped the GEO foundation on josecabal.com from scratch and I run it for clients across Metro Vancouver. What I am writing here is what I did, what worked, and what the research says about why. I will name every tool, link every number, and tell you what to ignore.

How big this actually is, in 2026

It helps to size the shift before talking tactics. ChatGPT crossed 900 million weekly active users in February 2026, more than doubling from 400 million a year earlier. Per Similarweb's January 2026 Market Research Panel, 35 percent of US consumers now use AI tools at the product discovery stage, compared to 13.6 percent who use traditional search for the same job. Google still owns most of the query volume across all device types, but the slice that AI answers absorb is concentrated in informational and research queries. That is exactly the space your blog and service pages compete for.

For the SMB I work with (a five to fifty person B2B services or manufacturing business in BC or south of the border), that is the slice that mattered most for organic lead generation. So the question is not whether to do GEO. The question is what to keep doing, what to start, and what to stop.

What carries over from SEO

This is the part most agencies will not tell you, because their pitch needs a clean break. The truth is messier. AI engines do not crawl the web from scratch. They lean heavily on the existing search index. Foundation Marketing's March 2026 analysis found that 61 percent of pages cited inside AI answers use structured data markup and 68.7 percent follow a logical heading hierarchy. Those are the same hygiene factors a competent SEO has been fighting for since 2018.

The technical stack that wins in AI search is the technical stack that wins in regular search. Fast pages. Clean HTML. Schema markup that actually validates. Internal linking that lets the crawler find every page in three clicks or fewer. A sitemap that is current. A robots.txt that does not accidentally block the crawler you want.

One nuance. AI search is more sensitive to recency than classic Google ever was for evergreen topics. Perplexity in particular weights freshness aggressively. If your content has not been touched in eighteen months, you are competing against pages that updated last week. That is not a new SEO principle, but the penalty for ignoring it has gone up.

GEO vs SEO: what actually changes

Dimension	Traditional SEO	GEO (Generative Engine Optimization)
What you are optimizing for	A blue-link ranking on the SERP	A passage citation inside an AI-generated answer
Primary success signal	Position 1–10, organic click	Named citation in ChatGPT, Perplexity, or AI Overviews
Content unit	The full page	A 134–167 word self-contained passage under a clear heading
Schema priority	Title, meta description, H1	FAQPage, Article/BlogPosting, Person + Organization entity graph
Freshness weight	Moderate (Google favours recent content for news topics)	High (Perplexity and AI Mode weight recency aggressively)
Backlinks	Critical for domain authority and ranking	Important but secondary to passage structure and entity clarity
Measurement	Impressions, clicks, position in GSC	Citation count in AI engines, AI-attributed sessions in GA4

The table above makes the difference look clean. It is messier in practice. Most of the work that wins in GEO is the same work that wins in SEO. The table is not a replacement plan — it is a priority reorder.

What is genuinely new

The interesting work is here. Three shifts changed the playbook this year.

Passage-level extraction is now the unit of competition

Google ranked pages. AI engines extract passages. A 134 to 167 word chunk under a clear heading, written as a self-contained answer to a specific question, is what gets lifted into a Perplexity citation or a ChatGPT answer. The page-level optimization frame still matters for indexing, but the user-facing result is a paragraph, not a result snippet.

Practical implication: write each H2 section so the first paragraph could stand alone if quoted. Lead with the answer. Put the qualifier and the example after.

llms.txt and the AI crawler split

There is a new file convention at the root of websites called llms.txt. It is a markdown file that tells AI assistants what your site is, what services you offer, and how to refer to your business. We ship one at josecabal.com/llms.txt as a worked example. Adoption is still low, no major model formally promises to honour it, but the cost is small and the dev-tool ecosystem (Cursor, Continue, Claude Code) already reads it.

The bigger change is in robots.txt. There are now two distinct categories of AI crawler. Search and retrieval bots (OAI-SearchBot, Claude-SearchBot, ChatGPT-User, Claude-User, PerplexityBot) fetch your pages to answer a user's live question with a citation. Training crawlers (GPTBot, ClaudeBot, Google-Extended, Applebot-Extended, CCBot) pull your content into the next model's training data. You can allow one category and block the other. Most B2B SMBs should allow the search bots and decide deliberately on the training bots based on their own IP and licensing posture.

Identity graphs as a first-class signal

AI engines try to resolve the entity behind a website before they cite it. Who is this person. What is their company. Where is it based. What do they do. The cleanest way to feed that resolution is a JSON-LD @graph in the static HTML that combines a Person node, an Organization or LocalBusiness node, and the relationships between them. We use the same pattern on this site. The identity graph sits in index.html and is therefore visible even to AI bots that do not execute JavaScript. That is the second non-negotiable for any SMB shipping a new site this year.

What is dead or fading

Three patterns I would tell a client to stop doing in 2026.

FAQ rich snippet farming

Google announced in August 2023 that FAQ rich results would only appear for "well-known, authoritative government and health websites." For everyone else, the rich result is gone. Search Engine Journal confirmed the August 8, 2023 cut-off. The implication is not that FAQ schema is useless. It is still cited by AI engines and still helps with passage extraction. The point is you should stop expecting the visual stars-and-arrows treatment in the SERP. The schema works as content structure, not as decoration.

HowTo schema

HowTo rich results were fully removed across desktop and mobile in the same announcement. There is no remaining rich result. There is no engagement benefit. Strip HowTo from your existing pages and stop generating it on new ones. It is bytes for nothing.

Broad keyword stuffing

Pages built around a head term ("AI marketing", "marketing automation", "SEO for SMBs") used to get away with thin content if they had enough internal and external links. AI engines do not extract from a broad term. They extract from the specific question a user asked. Pages over 20,000 characters in the ConvertMate 2026 benchmark averaged 10.18 AI citations each, compared to 2.39 for pages under 500 characters. That is a 4.3x multiplier on depth. Bad content is a tax on your brand. Thin content is a tax on your AI visibility.

A worked example: what we shipped on josecabal.com

To make this concrete, here is what we changed on this site over a single sprint to make it AI-search ready before launch.

Identity graph in index.html: a JSON-LD @graph with a Person node (José), an Organization+LocalBusiness node (the consultancy, with Vancouver coordinates and service area), and a WebSite node. Visible to AI bots that do not run JavaScript.
Per-route schema: BlogPosting on every article, BreadcrumbList on every interior page, Service on the offerings, FAQPage on the home page. Built from a single schema.ts module so the markup never drifts from reality.
Robots.txt with bifurcated AI crawler rules: explicit allow for OAI-SearchBot, Claude-SearchBot, PerplexityBot, ChatGPT-User, Claude-User, Perplexity-User. Explicit disallow for GPTBot, ClaudeBot, Google-Extended, Applebot-Extended, CCBot. Sitemap reference at the bottom.
llms.txt at the root: clean markdown, services and contact, how AI assistants should refer to the brand.
Build-time sitemap generator that reads the routes from siteConfig.ts and the blog posts and case studies from the data files, so adding a new article never leaves the sitemap stale.
Per-route prerendering via a post-build crawler so the per-route JSON-LD reaches AI bots that do not execute JavaScript.

None of those moves required a CMS migration or a content rewrite. They required someone who knew the difference between the schema that ranks and the schema that does not. That is the difference between a generic AI marketing pitch and a real implementation. For the full technical implementation of structured data, see the dedicated schema markup guide for B2B SMBs.

How an SMB should start

If you are a CEO or marketing lead at a 5 to 50 person business and you have read this far, here is the order of operations I would actually run on a 90-day engagement. I am not going to invent uplift numbers, because every site starts from a different baseline.

Weeks one to three. Foundation. Audit current schema, fix broken or missing markup, ship the identity graph, refactor robots.txt with the AI crawler split, write llms.txt, generate the sitemap from data sources. Block any AI training bots you have a policy reason to block. Allow the search bots.

Weeks four to six. Content shape. Rewrite the top five revenue-driving pages so each H2 section opens with a self-contained answer. Add genuine FAQ sections to the home, services, and pricing pages. Cite every claim with an inline link to a real source. Replace generic stock copy with marketer-specific detail: named tools, real benchmarks, real workflows.

Weeks seven to twelve. Authority and citations. Publish two or three depth posts per month at 1,800 words plus, each grounded in real data and primary observation. Earn mentions on Vancouver Tech Journal, BetaKit, or whichever local publication serves your audience. Get listed on Clutch, GoodFirms, and the directories your buyers actually check before a discovery call.

That is the boring fundamentals stack. It works because it stacks two things at the same time: being indexable by AI engines and being worth citing.

What this looks like in Metro Vancouver right now

If you run a B2B business in Metro Vancouver, the first-mover window is real and it is still open. Run a search in ChatGPT or Perplexity for "marketing consultant Vancouver," "AI marketing North Vancouver," or "fractional CMO Burnaby." The citation pool is thin. Most businesses that should appear there do not — because they blocked the search bots in robots.txt, they have no identity graph, and their pages have no FAQ schema. You can own that citation space with two weeks of technical work and eight to twelve well-structured pages.

The clients I work with in the Lower Mainland face one local wrinkle the generic GEO guides do not cover: the BC market has a shorter pipeline from AI citation to inbound call than the US market data suggests. Buyers here are fewer in number and more likely to do a single concentrated research session before reaching out. That makes passage clarity and contact information in structured data more important, not less. A business that shows up once with a clear, cited answer and a verifiable address in Burnaby or Richmond closes more of those citations than a US-market company that ranks broadly but vaguely.

What I actually see in the numbers

I track josecabal.com across a fixed set of 20 queries in ChatGPT, Perplexity, Claude, and Google AI Mode each month. In early tracking, Perplexity cites the site most consistently on Vancouver-specific queries; ChatGPT surface the site mainly when the query names Vancouver or the Lower Mainland explicitly. The citation count is growing as more structured content is indexed by AI engines.

What I can confirm from GA4: AI-referred sessions from chatgpt.com, perplexity.ai, and claude.ai are now a measurable share of organic traffic on this site. The volume is smaller than a top-5 Google ranking would produce. The intent is higher — visitors from AI citations have already had their question half-answered. They arrive to confirm, not to browse. That is a meaningful quality signal even before the numbers scale.

Frequently asked questions

What is the difference between SEO and GEO in plain English?

SEO optimizes a web page so it ranks high on a Google results page and earns a click. GEO optimizes the same page so an AI assistant (ChatGPT, Perplexity, Claude, Google AI Overviews) quotes it inside an answer with a citation. The work overlaps a lot. The unit of competition for GEO is a passage, not a page.

Should a Vancouver SMB still invest in traditional SEO?

Yes. Google still drives roughly 80 percent of global query volume per StatCounter and Similarweb, and 93 percent of pages cited inside AI Overviews are already in the top organic results for the same query. SEO is the foundation. GEO is the extension. Skipping SEO to chase AI citations is a strategic mistake.

How much of a B2B marketing budget should go to GEO in 2026?

For most SMBs in the 5 to 50 person range, GEO does not need a separate budget line yet. It needs the SEO budget to be redirected. The same agency or consultant or fractional CMO should be doing both, and the technical lift to add llms.txt, fix schema, and bifurcate the AI crawlers in robots.txt is measured in days, not months.

Does Perplexity favour recent content?

Yes. Perplexity weights freshness more aggressively than ChatGPT or Google AI Overviews, especially on topics where the underlying facts change quickly: tool pricing, model capabilities, regulatory news. Refresh evergreen pages every quarter and timestamp the update. A visible "last updated" line is a small change with a real signal.

Should I block GPTBot and ClaudeBot in robots.txt?

That is a business decision, not a technical one. Blocking the training bots prevents your content from being absorbed into the next model. Allowing them gives the model a richer view of your domain over time, which can (but does not always) translate to more confident citations later. Most of my clients allow the search and retrieval bots and block the training bots until they have a licensing conversation they want to have.

What is llms.txt and is it worth shipping?

llms.txt is a markdown file at the root of your domain that gives AI assistants a clean summary of your site, services, and brand voice. No major AI provider has formally committed to honour it. Adoption is in the early single digits. The cost is one file. We ship one on this site at josecabal.com/llms.txt on the principle that small low-risk infrastructure bets compound.

Do I still need FAQ schema in 2026?

Yes, but for different reasons. The FAQ rich result is gone for non-government, non-health sites since Google's August 2023 update. The schema itself remains useful. AI engines lift FAQ pairs into answers, and the structured Q-and-A format is the cleanest passage shape for extraction. Use FAQPage schema, but stop expecting visual rich snippets.

What is an identity graph and why does it matter for AI search?

An identity graph is a JSON-LD block that connects a Person, an Organization, a LocalBusiness, and a Website with stable @id anchors so AI engines can resolve who runs the site and how those entities relate. For a solo consultant or small team, this is the single biggest E-E-A-T signal you can ship in one file change. We put ours directly in index.html so non-JavaScript AI bots see it.

How long do AI citations actually drive traffic?

Less than a Google #1 ranking historically did, more than a social media post. The pattern I see with B2B clients is that AI citations drive a smaller volume of higher-intent visitors. The user has already had their question half-answered by the assistant, so the click is research-completion, not browsing. Conversion rates on AI-referred sessions tend to be higher than generic search sessions.

Will Google AI Mode kill organic traffic?

Not for branded and transactional queries. It will compress informational query traffic significantly. The defence is the same as the GEO playbook: depth, schema, original data, named-entity authority. Pages built for the user, not for a head keyword, survive.

What single change makes the biggest GEO difference for a B2B SMB?

Adding statistics with citations to existing service and pillar pages. The Princeton GEO study found up to 40 percent visibility lift from this technique alone. It is the lowest-effort, highest-payoff change on the list.

How do I measure if GEO is working?

Track three signals. One: branded mentions inside AI assistants. Ask ChatGPT, Perplexity, and Claude about your category every month and see if your name comes up unprompted. Two: direct or branded traffic from AI referral domains in Google Analytics 4. Three: organic CTR on ranked pages. When a brand is cited inside an AI Overview, organic CTR on the underlying page typically rises, not falls.

The short version

GEO is not a pivot away from SEO. It is the honest version of what SEO should have been all along: structured, cited, specific, written for a real reader rather than for a keyword. The agencies selling it as a clean break are selling a rebrand of their retainer. The marketers treating it as the next iteration of the same fundamentals are quietly accumulating citations while everyone else debates the acronym.

For a step-by-step guide to implementing GEO on an SMB site — covering the identity graph, schema, passage architecture, and AI crawler settings — see the GEO for SMBs guide.

The work is not complicated. The identity graph, the robots.txt split, the FAQ schema, the self-contained H2 sections — none of it requires a platform migration or a content agency. It requires someone who knows the difference between the schema that gets cited and the schema that does not. If you want that work done on your site, the services page is the place to start. Or if you want the full system end to end, the AI Marketing Boost covers it. The fundamentals compound. They always have.

Frequently asked questions

What is the difference between SEO and GEO in plain English?

Should a Vancouver SMB still invest in traditional SEO in 2026?

Yes. Google still drives roughly 80 percent of global query volume, and 93 percent of pages cited inside AI Overviews are already in the top organic results for the same query. SEO is the foundation. GEO is the extension. Skipping SEO to chase AI citations is a strategic mistake.

How much of a B2B marketing budget should go to GEO in 2026?

For most SMBs in the 5 to 50 person range, GEO does not need a separate budget line. It needs the existing SEO budget redirected. The technical lift to add llms.txt, fix schema, and configure the AI crawler split in robots.txt is measured in days, not months.

Does Perplexity favour recent content?

Yes. Perplexity weights freshness more aggressively than ChatGPT or Google AI Overviews, especially on topics where the underlying facts change quickly, such as tool pricing, model capabilities, and regulatory news. Refresh evergreen pages every quarter and timestamp the update. A visible last-updated line is a small change with a real signal. Content that has not been touched in eighteen months is competing against pages that updated last week.

Should I block GPTBot and ClaudeBot in robots.txt?

That is a business decision, not a technical one. Blocking the training bots prevents your content from being absorbed into the next model's training data. Allowing the search and retrieval bots (OAI-SearchBot, Claude-SearchBot, PerplexityBot) is what drives AI citations. Most clients allow search bots and block training bots until they have a licensing conversation they want to have.

What is llms.txt and is it worth shipping?

llms.txt is a markdown file at the root of your domain that gives AI assistants a clean summary of your site, services, and brand voice. No major AI provider has formally committed to honour it, but the cost is one file. Early adoption is a low-risk infrastructure bet that compounds.

Do I still need FAQ schema in 2026?

Yes, but for different reasons. The FAQ rich result is gone for non-government and non-health sites since Google's August 2023 update. The schema itself remains useful because AI engines lift FAQ pairs into answers, and the structured Q-and-A format is the cleanest passage shape for extraction. Foundation Marketing's March 2026 analysis found that 61 percent of pages cited inside AI answers use structured data markup. Use FAQPage schema, and stop expecting the visual rich snippet in the SERP.

What is an identity graph and why does it matter for AI search?

An identity graph is a JSON-LD block that connects a Person, an Organization, a LocalBusiness, and a Website with stable @id anchors so AI engines can resolve who runs the site. For a solo consultant or small team, this is the single biggest E-E-A-T signal you can ship in one file change.

Will Google AI Mode kill organic traffic?

Not for branded and transactional queries. It will compress informational query traffic significantly, because that is the slice AI answers absorb first. The defence is depth, schema, original data, and named-entity authority. Pages written for a real reader survive, and pages built around a head keyword do not. In the ConvertMate 2026 benchmark, pages over 20,000 characters averaged 10.18 AI citations each, compared with 2.39 for pages under 500 characters.

How long does a GEO foundation take to ship for a typical Vancouver SMB site?

Three weeks for a 30–80 page site. Week one is the audit: existing schema, robots.txt, sitemap, llms.txt absence, AI crawler coverage. Week two is the technical fixes — identity graph in the HTML head, FAQPage schema on the top 10 pages, BreadcrumbList everywhere, robots.txt rewrite. Week three is content shape: rewriting the top 5 H2 sections on the highest-traffic pages so each opens with a self-contained answer. Beyond that, the work becomes ongoing content production, not one-time setup.

Does GEO work for B2B SaaS, professional services, and local services equally?

Yes, with different topic surfaces. B2B SaaS earns the most citations on integration and comparison queries (HubSpot vs Salesforce, Claude Sonnet 4.6 vs GPT-5). Professional services earn citations on pricing breakdowns and process explanations. Local services earn citations on neighbourhood-specific queries (Mount Pleasant SEO, Burnaby commercial cleaning). The schema and FAQ structure is the same. The query targeting is different.

How much does a GEO audit cost in Canada in 2026?

A scoped GEO audit for a 5–50 person SMB runs CAD $1,500–$4,000 depending on site size and existing schema state. The deliverable is a prioritized fix list with the top 10 page-level changes, the robots.txt and llms.txt patches, and the FAQ structures to add to the top revenue pages. The implementation lift after the audit is typically 15–30 hours of developer and content time.

What is the single biggest mistake SMBs make when starting GEO?

Treating it as content volume. Publishing 50 thin posts to chase AI citations is the velocity-era playbook and it backfires in 2026. The Helpful Content System penalises domains with too much thin content, and AI engines cite passages from authoritative pages, not from page count. Ship 8–12 deep posts a quarter with FAQ schema, citations, and named entities. Cite the source on every data point. That is what gets quoted in ChatGPT.

How would I prove a GEO investment paid off?

Three numbers tracked monthly. Citation count in ChatGPT, Perplexity, Claude, and Google AI Overviews against a fixed 20-query manual sample. Branded mention volume in AI assistants (ask each engine 'who does AI marketing in Vancouver' once a month). AI-attributed sessions in GA4 using a custom channel group matching chatgpt.com, perplexity.ai, gemini.google.com, claude.ai, and copilot.microsoft.com referrers. Pipeline contribution from those sessions in HubSpot. If those climb over a 90–180 day window, the work paid off.