SEO, GEO, CRO & demand engineering
Generative engine optimisation, measured with evidence rather than vendor scores.
In short
VenSoc runs technical SEO, generative engine optimisation (GEO) and conversion rate optimisation (CRO) programmes for B2B companies. GEO makes a site retrievable and citable by AI answer engines: ChatGPT, Claude, Perplexity, and Google AI Overviews. CRO is the work that follows, on what visitors do once they have arrived. Both are measured with first-party data rather than vendor scores.
- Disciplines
- Technical SEO, GEO, CRO, content architecture, measurement
- Primary metric
- Verified AI crawler fetches from your own server logs
- Secondary metric
- Prompt-panel citation rate, 4-week trailing mean
- Reporting cadence
- Monthly, with raw data included
- Engagement
- Audit, then retained programme
What this includes
- Technical SEO audit
- Generative engine optimisation
- AI crawler log analysis
- Content architecture for retrieval
- Structured data implementation
- Prompt-panel measurement
- Conversion rate optimisation
- Analytics and funnel review
- A/B testing
- Checkout and landing-page optimisation
What actually causes an AI engine to cite a source?
The strongest published predictor of AI visibility is off-site brand mention volume, not backlinks. An analysis of 75,000 brands found YouTube mentions correlating at r≈0.74 and branded web mentions at r≈0.66–0.71, against r≈0.22 for backlinks. On-page work matters, but it is the smaller lever.
The second finding that changes strategy: AI engines cite a substantially different web than Google does. A University of Toronto study across 1,516 queries and five systems found only 4–15% domain overlap between Google's top results and AI citations. Ranking well is neither necessary nor sufficient for being cited.
The third: citations are volatile. An analysis of 1,127 URLs over 28 days found citation persistence dropping to 10.6%, with under 17% overlap in cited domains across platforms. Any measurement design that samples once is measuring noise. VenSoc reports four-week trailing means and never a point-in-time visibility score.
Why does the technical foundation matter more than it used to?
No major AI crawler executes JavaScript. Analysis of roughly 1.3 billion crawler fetches found GPTBot requesting JavaScript in 11.5% of cases and never executing it, and ClaudeBot in 23.8% of cases, also never executing it. Content rendered client-side is invisible to the entire channel.
This single fact voids a large amount of otherwise competent GEO work. A site built as a client-rendered application can hold excellent content, perfect structured data and a flawless llms.txt, and still be uncitable, because the crawler receives an empty shell.
VenSoc audits this first, by fetching each route with an AI crawler user-agent and checking whether the sentences the client wants quoted are present in the raw HTML response. On vensoc.com the same check runs in continuous integration and fails the build.
What does VenSoc refuse to sell as a GEO tactic?
VenSoc does not sell llms.txt as a citation driver, does not sell schema markup as an AI ranking lever, and does not sell any third-party AI visibility score as a metric. All three are widely marketed. None of them survives contact with the evidence.
- llms.txt
- Google described the standard as "purely speculative for now" in June 2026. Server-log analysis recorded 84 of 62,100+ AI bot requests touching it. Worth shipping — it costs half an hour and carries no risk — but it is not a lever.
- Schema markup for AI citation
- No measured correlation between schema coverage and citation rate. Schema is valuable for Google rich results and entity disambiguation. Sold as a GEO tactic, it is misrepresentation.
- FAQ schema for search visibility
- Google removed FAQ rich results from Search on 7 May 2026 and ended reporting support in June. The markup remains parseable by LLM pipelines; it produces zero Google visibility.
- Vendor AI visibility scores
- With citation persistence at 10.6% over four weeks and under 17% cross-platform overlap, a single-sample score is not a measurement. VenSoc runs a fixed prompt panel across multiple engines weekly and reports the trailing mean.
What does conversion rate optimisation involve, and when is an A/B test worth running?
Conversion rate optimisation is three pieces of work: an analytics and funnel review that establishes where visitors actually leave, controlled A/B tests of the changes that review suggests, and direct improvement of checkout and landing pages. The review is diagnostic, the testing is evidence, and the direct work is where the largest single gain usually sits.
- Analytics and funnel review
- Where visitors leave, established from the analytics already in place rather than from an opinion about the design. This step frequently finds a broken step rather than a persuasion problem.
- A/B testing
- A controlled comparison, run only where the traffic can support a conclusion. Where it cannot, VenSoc says so rather than reporting a difference indistinguishable from noise.
- Checkout and landing pages
- The pages where intent is highest and friction is most expensive: form length, payment steps, error handling, and what the page asks for before it has earned it.
Common questions
- What is the difference between SEO and GEO?
- SEO optimises for ranked lists of links; GEO — generative engine optimisation — optimises for being retrieved and quoted inside a generated answer. The mechanisms differ: search ranks pages, while AI systems chunk pages, embed the chunks, and retrieve at passage level. A page can rank first in Google and never be cited, and vice versa. In practice the two share a technical foundation and diverge sharply on content structure and on how success is measured.
- How do you measure whether GEO is working?
- Three sources. First, your own server logs, parsed for the eleven AI crawler user-agents and verified by reverse DNS — this is the only first-party, non-inferential GEO metric that exists. Second, referral traffic segmented for chatgpt.com, perplexity.ai, claude.ai and gemini.google.com; expect roughly 1–3% of total traffic. Third, a fixed panel of 50–100 buyer prompts run weekly across multiple engines, reported as a four-week trailing mean.
- Should we block AI crawlers from training on our content?
- For most B2B services businesses, no. The distinction that matters is that training bots and retrieval bots are different agents: blocking GPTBot does not remove you from ChatGPT answers, whereas blocking OAI-SearchBot does. If you have no content moat to protect, presence in the training corpus is how your company gets named from model memory rather than requiring live retrieval. Publishers and businesses whose content is the product should reach the opposite conclusion.
- How long before GEO work shows results?
- The technical foundation — crawlability, structure, structured data — can be fixed in weeks and is a prerequisite for everything else. Content and earned-mention work operates on a three to nine month horizon, because the dominant signal is off-site mention volume, which accumulates rather than switching on. Any agency promising AI citation gains in weeks is describing something they cannot control.
- How much traffic does a site need before A/B testing tells you anything?
- It depends on the baseline conversion rate and the size of the change being tested, not on a traffic threshold anyone can quote. A B2B site with a small number of qualified visits per month will often never reach significance on a realistic effect size, and running a test anyway produces a number that looks like evidence and is not. VenSoc says so before the work starts, and spends the budget on funnel review and direct checkout and landing-page work instead.