What are the common mistakes marketing agencies make when starting GEO?

Q: What is the single most common GEO mistake?

Skipping the baseline scan. Without a before-picture across ChatGPT, Perplexity, Gemini, Claude, and DeepSeek on a fixed prompt set, there is no defensible way to prove movement in month four. This single omission kills more renewals than any content or technical mistake.

Q: How many prompts should a tracked set include?

20-40 prompts across branded, category, and adjacent question types. Scan weekly, report monthly. The exact count matters less than the consistency of the cadence.

Q: Can I reuse my SEO content playbook for GEO?

Not without significant changes. SEO content optimises for ranking; GEO content optimises for being lifted into an AI answer. The brief, structure, and editorial review steps are different. Copy-pasting the SEO playbook produces pages Google might rank but AI engines will not cite.

Q: How important is JSON-LD schema for GEO?

It is the biggest accelerator on first-citation timing in our experience. Pages with valid Article, Organization, and FAQPage schema get cited earlier and more often. Skipping the schema work to focus on content alone is a common but costly mistake.

Question

What are the common mistakes marketing agencies make when starting GEO?

Maciej Grabek · Accepted Answer

We have watched a lot of agencies start GEO programs and stumble on the same five mistakes: no baseline scan, no tracked prompt set, copying SEO content playbooks verbatim, overpromising fast results, and skipping the boring JSON-LD schema work. None of these are agency-killers individually. Together, they make the first three months of a program look flat, which is when most retainers get cancelled. Mistake 1: No baseline scan before work starts The agency wins the deal, fires up content, and three months later cannot prove anything moved because there is no before-picture. Citation share on month four reads "we are at 12%" and the client asks "compared to what?" The fix is mechanical: before any work starts, scan all five engines (ChatGPT, Perplexity, Gemini, Claude, DeepSeek) on the client's tracked prompt set. Save the raw responses, save the citation lists, save screenshots. That is the baseline. Every QBR references it. We have seen agencies skip this for 30+ engagements before getting burned on a renewal. Mistake 2: No tracked prompt set Closely related, but worse. Without a fixed list of prompts scanned on a regular cadence, there is no signal. The agency reports "we got cited in ChatGPT this month" anecdotally, but cannot show frequency, share, or trend. Clients lose patience by month two. A workable prompt set is 20-40 prompts across three categories: branded ("is [client] a good [category] tool?"), category ("best [category] tools for [use case]"), and adjacent ("how do I solve [problem [client] solves]?"). Scan weekly. Report monthly. This single discipline separates agencies that retain clients from agencies that churn at six months. Mistake 3: Copy-pasting the SEO content playbook SEO content is optimised for Google's ranking system: keyword density, backlinks, header structure, dwell time. GEO content is optimised for being lifted into an AI answer: entity density, opinionated framing, concrete examples, JSON-LD schema, and structured Q&A. An e-comm-focused growth shop that writes 1,500-word "ultimate guides" stuffed with keywords will produce pages Google might rank but ChatGPT will not cite. The pages are too generic, the entities too vague, the answer too buried. We have seen this play out at agencies of every size. The fix is editorial. Brief writers to lead with the answer, name brands and engines specifically, include real prompts and excerpts, and structure for liftability (TL;DR up top, scannable H2s, clear Q&A blocks). Mistake 4: Overpromising The temptation is real. The buyer wants to know what they get for their money, and "citation share will move on a 60-120 day curve" is a slower close than "we will get you to the top of ChatGPT in 30 days". Agencies bend, promise the latter, and have nothing to show at the 30-day check-in. The recovery is brutal: re-set expectations, lose trust, lose the renewal. The cheaper move is to set realistic expectations up front, anchor on tracked-prompt movement instead of vague "top of ChatGPT" outcomes, and over-deliver against the lower bar. Mistake 5: Skipping the schema work JSON-LD and llms.txt are not glamorous. They do not generate slides for a QBR. But they are the single biggest accelerator we have seen on first-citation timing. Pages with valid Article, Organization, FAQPage, and (where applicable) Product schema get cited earlier and more often. Agencies skip this either because the team does not know how, or because it sits awkwardly between marketing and engineering. Either way, every week without proper schema is a week of slower compounding. Fix it in week one. Re-audit quarterly. The pattern behind all five All five mistakes share a root cause: treating GEO like SEO with a new audience instead of a new discipline with its own measurement, content, and technical playbook. Agencies that get past the first 90 days successfully are the ones that build the GEO muscle as a distinct capability, not as an SEO bolt-on. The agencies that struggle are the ones that send the SEO lead to a webinar and call it a strategy. None of this is hard. It just requires deciding GEO is a real thing, then doing the work it needs.