Questions to Ask Before Hiring a GEO Agency (Written by One)

If you are vetting a generative engine optimization (GEO) agency, seven questions do most of the filtering: what baseline will you set, what happens in the first 30 days, which engines do you track, where do my third-party mentions come from, how does this fit my existing SEO, what will you promise and by when, and who do you turn away. The same list applies whether the firm calls the work GEO or answer engine optimization (AEO), because the two labels describe [one discipline](/blog/geo-vs-aeo). The hub guides cover the category definitions properly, so I will point you at [what generative engine optimization is](/blog/what-is-generative-engine-optimization) and [what answer engine optimization is](/blog/what-is-answer-engine-optimization) rather than re-teach them here.

One disclosure before the list. Lore is a [GEO agency](/blog/geo-agency-vs-diy). That could make this piece a brochure, so I have written it the opposite way: these are the questions I would want a buyer to grill us with, including the ones a vendor in our position enjoys least. If a firm squirms at any of them, you have learned something worth more than the proposal deck. And if you want the plain description of what you would be buying first, the [breakdown of what a GEO agency does](/blog/what-a-geo-agency-does) sits one step before this article.

The seven questions, with good and bad answers

### 1. What baseline will you set, and how will you measure movement against it?

**A good answer** describes a month-one audit of how ChatGPT, Perplexity, Gemini, and Google's AI Overviews currently respond to the commercial prompts in your category: whether you are mentioned, whether you are cited, who the standing recommendation is, and which sources the engines lean on. Every later report gets measured against that snapshot, and every number arrives paired with its comparison window.

**A bad answer** is any version of "we'll get you more AI mentions" with no snapshot to compare against. Mentions without a baseline are a vanity feed. You are buying movement, and movement needs a starting point.

### 2. What happens in the first 30 days?

**A good answer** starts with foundations: crawler access, indexing, site speed, internal linking, schema. A frequent live finding when clients arrive on our GEO plans is AI crawlers blocked by default in robots.txt, usually a CMS preset nobody revisited. It is a cheap fix with a severe cost, and an agency that checks for it before writing a word of content has the sequence right.

**A bad answer** jumps straight to a content calendar. A brilliant editorial plan aimed at a site the engines cannot crawl reaches no one, and a vendor who skips the plumbing is billing you for output rather than outcomes.

### 3. Which engines do you track, and how do you report share of voice?

**A good answer** names the engines, names the prompt set, and commits to a cadence: how each engine answers your buyers' questions, tracked monthly, reported as mentions, citations, and share of voice against named competitors. Ask to see a real (anonymised) monthly report before you sign. The format tells you how they think.

**A bad answer** leans on a single dashboard screenshot or a proprietary "AI visibility score" that cannot be traced back to prompts you could run yourself. Opaque scores exist to make every month look like progress.

### 4. Where will my third-party mentions come from?

Every business says it is the best, so the engines weight independent corroboration heavily: third-party sources explaining what you do and who you serve are what shift which brands get cited. That makes this question a load-bearing one, and it has a specific honesty test inside it.

**A good answer** admits that purely earned links are too slow to promise for a brand without existing gravity, and explains how paid press distribution fits the standard package so link outcomes are deliverable, alongside genuine participation in the communities the engines draw on. That is the honest version of the trade.

**A bad answer** promises a specific count of purely earned mentions on a schedule, or waves at "outreach" with no named channels. Either the count will quietly become paid anyway, or the schedule will slip and the reporting will blur why.

### 5. How does this fit with the SEO I am already running?

**A good answer** refuses the binary. Traditional search still carries serious volume in most categories, AI surfaces are growing on top of it, and work done correctly serves both at once: one programme, two reporting surfaces. The backdrop we publish on our own site is that organic search traffic is down roughly 30% across the board while AI referrals convert at roughly 3 to 6x the rate of traditional organic visitors, which argues for adding a surface rather than abandoning one.

**A bad answer** is any pitch built on classic search being finished. A firm selling you panic about the channel you already own will sell you panic about the next channel too. If budget is part of this conversation, the [honest numbers on what GEO costs](/blog/how-much-does-geo-cost) are worth reading before any sales call.

### 6. What will you promise, and by when?

**A good answer** sounds unimpressive in a pitch meeting, which is exactly why it is trustworthy: early technical wins inside the first month, then compounding movement in mentions and citations over one to two quarters, because citations follow crawl cycles, index refreshes, and model updates that no vendor controls. In full transparency, this work takes months to snowball, and we say that to clients before they sign rather than after.

**A bad answer** guarantees named placement in AI answers by a fixed date. Nobody controls what a model composes next month. If a vendor promises guaranteed AI citations in weeks, walk. That single sentence filters more of this market than the other six questions combined.

### 7. Who do you turn away?

**A good answer** has real examples. An agency with a genuine qualification ethic can describe the prospects it declined and why: markets too small, ceilings too low, timelines the buyer could not fund. We hold the position that you only take clients who can win, because selling a programme to a business that cannot succeed from it helps nobody and eventually shows up in your own case studies.

**A bad answer** is a pause followed by "we can help anyone." A firm with no disqualification criteria has one criterion, which is your signature.

The questions a vendor like us enjoys least

Three more, for when you want to see how a firm behaves under mild discomfort. Ask to see a prompt set they failed to move and what they changed afterwards. Ask which of their reported metrics are proxies rather than direct measurements, and how they flag the difference. Ask to speak to a client whose programme took longer than projected. None of these have crowd-pleasing answers, and that is the point: a vendor who handles them plainly will handle your bad months plainly too. When we publish receipts, Tides Mental Health at 4.7x organic traffic growth in five months or BizScout with 450+ AI mentions built across a year, each figure sits inside a programme description and a comparison window, because numbers stripped of their context are how this industry misleads people.

When GEO is not worth buying yet

This is the section most agencies would leave out, and it is the strongest filter you can run before any vendor call, because the best question on this list is one you ask yourself.

**The ceiling is too low.** A hyper-specific single-product business, say a workshop selling one specialised bench for one instrument, may have a winnable market too small to repay a retainer no matter how well the work goes. The spend has to be justified by what winning is worth. A decent agency prices that in the first conversation and declines when the maths fails; we turn that work down for exactly this reason.

**You have the operator and the scope for DIY.** A technically capable in-house marketer, a narrow prompt set, and a content engine that already ships can get meaningful distance from a monitoring tool plus internal hours. Buy the retainer when the workstreams outgrow the operator.

**The budget cannot survive the compounding window.** If the business needs leads next month, paid has a genuine place and is often the right first spend. GEO pays on the timescale of quarters, and funding it out of money that was needed this month turns a compounding asset into a source of panic.

If two or more of those describe you, bookmark this and come back in six months. The market will still be here, and the vendors worth hiring will respect you more for the timing.

One outside voice worth your time

For a substantive third-party view of what separates a real programme from a rebadged SEO retainer, the Ahrefs video "Top SEO Experts Build Me an AI Search Strategy (GEO)" is one of the better watches available. It shows experienced practitioners building an AI search strategy from scratch, and seeing the choices they weigh helps you judge whether a vendor across the table knows the same mechanics.

<YouTube id="RwKKLnyXCig" />

The short version

Ask for the baseline, the first-30-days sequence, the engine list, the source of your mentions, the SEO relationship, the promise with its timeline, and the clients they refused. Good answers arrive with mechanisms and comparison windows attached, while the bad ones lean on adjectives and guarantees. And if your ceiling, your operator, or your budget says wait, then wait.

If you want to run the vetting on us first, download the [46-point AI visibility checklist](https://lorebuilders.com) and score any proposal you receive against it, ours included. If the scores are close, a strategy call will settle the rest.