Seer Published 27 Questions to Ask a GEO Agency. Here Are Our Answers, Including the Ones We Fail

Wil Reynolds published 27 questions a buyer should ask any agency pitching GEO. We answered all of them in public.

We fail four of them.

As far as we can tell, no agency has published its own answers to that guide, which is the only reason this page is worth writing. A buyer in Riyadh or Cairo can now put these answers next to whoever else is pitching, and compare like with like.

The questions come from Seer Interactive’s GEO RFP guide, published on 10 August 2026 and built from hundreds of their own sales calls. They are paraphrased and grouped below. Read the original for Reynolds’s reasoning, which is better than any summary of it.

Key takeaways

• Twenty-three of the 27 questions are answered here with evidence. Four are answered with a gap.
• The strongest proof is Delta Medical Labs: 5,570 pages cited by AI engines as of July 2026, with a four-engine breakdown.
• The weakest answer is question one, where the win exists but a documented 90-day follow-up on it does not.
• No published experiment with a rerunnable method and a named author. That is the biggest hole in our credibility.
• Numbers that go the wrong way get published anyway. The bounce rate in our Ogaei migration write-up rose 13.5% and it stayed in.

Where we would score badly

Four answers first, because a page like this is worthless if it only contains the flattering ones.

No 90-day follow-up on our best GEO result. Reynolds asks for the win, then for what happened at 30, 60 and 90 days. That second half is where GEO case studies usually fall apart.

Our Delta Medical Labs numbers are a July 2026 measurement, and there is no published follow-up measurement against the same prompts. Until one exists, treat the figure as a snapshot rather than a trend.

No rerunnable experiment. We have client outcomes. What we do not have is a public document carrying a falsifiable hypothesis, a sample size before and after cleaning, the tools, the dates and the limitation. Written so a stranger could repeat it.

Agencies that hold that document are ahead of us on question five, and the honest response is to say so rather than dress an outcome up as research.

No published null result. The experiments that disprove your own assumption are the most credible thing an agency can publish. We have not done it once.

One of our largest clients has no AI-citation data at all. FBS is a 6.5 year forex engagement across Saudi Arabia and the UAE, with 5X organic session growth.

It carries no AI-citation figure, because nobody measured one. Any agency page claiming AI citations for a financial services client of that size is inventing them. We would rather tell you a number does not exist.

What does a GEO win look like when it works?

Delta Medical Labs is a Saudi medical laboratory, engaged since March 2026. As of July 2026, 5,570 of its pages are cited by AI engines.

The breakdown matters more than the total, because it tells you which engine is doing the work. Google AI Overviews accounts for 3,600, AI Mode for 1,800, ChatGPT for 130 and Gemini for 40.

Third-party benchmark from the same month, via Semrush AI Visibility: Delta scores 18. Al Borg Diagnostics 14. Wareed 14. Alfa Lab 0. The only Saudi health entity scoring higher is the Ministry of Health at 22.

We publish the tracked sample as well as the total, so the number can be audited. That sample is 326 keywords tracked monthly on Google.com.sa, of which 197 sit at position one and 240 in the top three.

This is a health client, in Arabic, in a YMYL category. That is the hardest combination we work in, which is why it is the example we lead with.

How do we decide whether SEO is still worth your money?

We build the arithmetic before the argument.

Traffic trend, value per visit, conversion value, and the cost to run the programme are the four inputs. Together they produce a revenue model that tells you whether the spend still clears.

Reynolds makes a point most GEO agencies avoid, which is that SEO is still working, and our own roster says the same. FBS has run 6.5 years across Saudi Arabia and the UAE with 5X organic session growth.

Its last reporting cycle showed 164,400 organic clicks and 2.8 million impressions, with clicks up 50.7%.

An agency that tells you SEO is finished is selling the thing it wants to sell, rather than answering the question you asked.

Speed against risk, and the changes we will not rush

We separate reversible changes from irreversible ones, and we tell you which is which before making them.

Editing a cited page is reversible in an afternoon. A domain migration is not.

When Ogaei moved to a new domain and a new CMS in a single cutover, every improvement was frozen until the platform was proven stable. The 37-item checklist we used is published in full. The improvements shipped afterwards. That sequence is slower, and it is the reason nothing broke.

You will not get a promised ranking or citation rate from us. What you will get is a list of mechanics that are either correct or not: redirects resolving in one hop, no non-301 status codes, index coverage stable, crawl rate steady.

What are we testing right now, without knowing the answer?

Three things, and none of them has a conclusion yet.

Arabic against English citation behaviour, which almost nobody has data on. Most of our portfolio is Arabic, which is the only reason we are placed to measure it.

Second, whether Arabic register changes citation at all, meaning the same query asked in Gulf colloquial and then in Modern Standard Arabic.

Third, whether AI engine citation is stable enough run to run in Arabic to be measured at all. If it is not, the first two questions are unanswerable and we would have to say so. The interim findings are published, including why none of the 37 named GEO experts publishes in Arabic.

The honest position today is collecting rather than concluding. When it publishes, it will carry a method you can rerun and a limitations section written in plain words.

Where our prompt list comes from

From your customers, not from a keyword tool.

Call transcripts, paid search query data, the objections your sales team answers every week, and the questions your support inbox repeats.

Reynolds is right that agencies skipping this step end up building listicles, because a listicle is what you produce when you have no idea what your buyer actually asks.

Branded and non-branded prompts get split early. If someone asks an AI engine about your company by name and your own pages are not the source, another site is writing your description for you. That gets fixed first, and it is the less exciting half of the work.

What we do on your site, and what we will not do

The pages already being cited get optimised before any new ones are written, because a page an engine already retrieves is the cheapest thing to improve.

Original data goes in wherever you have it. Most companies sit on numbers nobody outside the company has seen, and unique data is the one thing a model cannot lift from your competitor’s page instead.

On converting your whole site to plain Markdown for agents, the answer is no. We have not seen evidence it helps enough to justify what it costs a human reader. Ask us to test it on one section and we will, then publish the result either way.

On volume, ten pieces against a hundred should not cost ten times more from an agency that has built its AI workflows properly. Two to three times is closer to honest. A pitch that scales linearly with volume is billing you for headcount.

Connecting AI visibility to money

The first thing worth saying is that visibility is not traffic and should not be reported as traffic.

Profound published panel research showing that after an AI assistant mentions a brand, users visit that brand’s site at 1.5 to 2.5 times their forecasted baseline over the following seven days, without clicking the citation.

That behaves more like a billboard than a blue link, which means reporting it as sessions understates what it did.

Scope discipline applies to every number we hand you. Our Ogaei write-up published GA4 session figures and Search Console click figures in the same article and labelled them separately, because they cover different windows and count different things. Mixing them is how agencies inflate results by accident.

Ovasave is the number we would put in front of a finance director: 46% growth in SEO conversions over six months. A conversion figure rather than a traffic figure, and therefore harder to argue with.

When has AI visibility not driven business results?

When the mentions land on informational queries and the buyer was never in the conversation.

There is a version of this published against ourselves. In the Ogaei migration write-up organic sessions grew 693.3% year over year, and in the same window the property-wide bounce rate rose 13.5%, to 47.6%.

We published the number that went the wrong way. Our reading is a traffic-mix shift, and that reading is labelled as a reading rather than a measured attribution.

A single blended engagement metric during rapid traffic growth is close to meaningless. Any agency quoting you one without a segment behind it is either not looking closely, or hoping that you will not.

How mature is our own AI operation?

This is the section most agencies would skip, and it is also the only one we can answer with artifacts rather than adjectives.

Our own work runs through documented skills rather than ad hoc prompting. There is a house voice profile with numeric targets, a style scorer run against every draft before it ships, and a proof bank that every claim of experience must map to. A number with no source in that file does not go on the site.

Experiments run on our own site before they run on a client’s, and three recent ones are worth naming.

A cannibalization audit across our own agency pages, using Search Console query and page data, found four of our own URLs splitting impressions on a single query. All four sat past position forty.

A full internal link status check found a broken call to action on our own site that had been live in three published articles.

A primary-source audit of OpenAI’s advertising documentation produced three findings that the secondary coverage had missed entirely. One of them, on what ChatGPT ads actually cost in the Arab world, contradicts what most published guides still say.

Corrections get published too. When our own migration article overstated that URL paths were held identical everywhere, we found the exception in the client’s own mapping sheet and rewrote the claim.

Keeping up when the engines change weekly

Standing weekly checks, rather than reactive scrambling.

One scheduled weekly review covers AI search and GEO changes angled at the Gulf, Egypt and the wider Middle East, and a separate weekly check covers OpenAI’s advertising availability and policy pages.

Those pages move fast enough to catch out anyone checking quarterly. OpenAI’s country availability went from 9 markets to 52 in under three weeks, and its billing documentation had been updated 11 hours before we last read it.

When a client’s numbers move, the first question is whether the engine changed its layout or its citation rate, not whether the work stopped working. Knowing that in advance is the difference between an explanation and a scramble.

Twenty-three answered with evidence. Four answered with a gap. That ratio is worth more published than a page claiming twenty-seven.

Objections we hear

“Every agency claims to be honest. Why should this page count?”

Because it names four things we cannot do and one client where a number does not exist, and marketing pages do not usually contain either of those.

Check the parts you can check for yourself. The Semrush benchmark is third-party and dated, and the Clutch rating is 5.0 from 8 reviews as of 17 August 2026. Every case study named here carries the client’s own name.

“You are in Egypt. We are in Saudi Arabia or the Gulf.”

Most of our portfolio is Arabic and Gulf-facing, which is easier to show than to claim. Delta is a Saudi laboratory, FBS ran 6.5 years across Saudi Arabia and the UAE, and Al Hokail is a Saudi medical group.

The question is not where the office sits, but whether the agency has done the work in your language and your market, and you should ask that of everyone you shortlist.

“Twenty-three out of twenty-seven still sounds like a sales page.”

Then use the four failures as your filter, and put the same four questions to every agency on your list.

Show me a 90-day follow-up on a GEO win. Show me a rerunnable experiment. Show me a published null result. Show me a client where you told them a number does not exist.

If they answer all four with evidence, they are ahead of us and you should hire them.

The more useful question

The question is not which agency has the best answers to these 27. Everyone will improve their answers now that the questions are public.

The question is which agency tells you what it cannot do before you sign, because that is the same agency that will tell you when something is not working after you sign.

If you are shortlisting a GEO partner in the Gulf or Egypt, take Reynolds’s questions into every meeting, including ours. Ask us for an AI search audit and we will show you which answers you already appear in and which ones your competitors own. We will also tell you which of these 27 we would still struggle with on your account.

Related guides

• Top 3 AI SEO Agencies for GEO & AEO in 2026: Who Actually Delivers Visibility Inside AI Answers
• What Does a GEO Agency Do? The Complete 2026 Guide
• Best GEO Agency in Egypt 2026: How to Choose and What to Expect
• Best AI Search & GEO Agency in the Middle East, Gulf & Egypt (2026)