Key takeaways
- Many agencies selling GEO are standard SEO services with a new label. Questions about measurement are the fastest way to tell the difference.
- Ask for the prompt list, the engines tracked, and the retest cadence in writing. A fixed prompt set tracked over time is the baseline for any credible answer.
- Reporting should separate citations, brand mentions, and recommendations, and it should show each AI engine on its own.
- Be wary of guaranteed rankings in ChatGPT. AI attribution is messy, so a month-by-month plan and a short pilot beat a promise.
- Make sure you keep the pages, prompt list, and raw data if the relationship ends.
I've sat through enough agency pitches to know how this goes. The deck says "GEO" on the cover, then slide six is a keyword ranking chart. This guide gives you twelve questions that expose that gap, grouped by what they test: tracking, reporting, and results. Each has a strong answer to listen for and a red flag.
A note on sources. Most public advice on this topic comes from agencies and vendors that want your business, so treat their retainer figures and checklists as self-reported. Several published checklists (5WPR's twelve-item list, ScaleTheory's buyer's guide, Percepture's SEO interview) agree on the basics, and I've folded those in with Promptwatch's first-party data on how AI engines behave.

Before the first call: do some homework
Two checks take ten minutes and save you an hour of pitch.
First, ask ChatGPT, Perplexity, and Gemini who the best agencies are in the agency's own category. If a company selling AI visibility cannot show up in AI answers for its own niche, that tells you something. ScaleTheory's buyer's guide makes the same point.
Second, look for published methodology. A GEO agency that has written down how it works is easier to hold accountable than one that says "proprietary process."
Part 1: Tracking questions
Tracking is where agencies most often get vague. These four questions force specifics.
1. Which prompts will you track, and who writes the list?
A good answer: a fixed set of prompts built from your buyers' real questions, shared with you in full, with a named person responsible for it. Prompts should cover branded, competitor comparison, and plain informational queries.
Red flag: "We track your brand across AI." Without a prompt list there is nothing to audit. One Reddit thread on this topic reduced it to a simple test: good means share of voice across a fixed prompt set, tracked over time.
2. Which AI engines do you measure, and how do you collect the data?
Ask for the list by name: ChatGPT, Gemini, Perplexity, Claude, Google AI Overviews, Google AI Mode, Microsoft Copilot. Then ask whether the data comes from the real interfaces people use or only from API calls. User-facing answers and citations can differ from API output.
Engines differ in how they cite, which is why per-engine reporting matters. Promptwatch Data's average sources per response report shows ChatGPT citing about 5 sources per web-search answer, while Google AI Overviews and Perplexity cite about 10. Copilot swung from under 2 to nearly 17 sources per response within a few weeks before settling. An agency that blends all of that into one number is hiding information.
3. How often do you retest, and what happens after a model update?
ChatGPT's search behavior changes in steps. Promptwatch's query fanout data shows average fanout query length falling from about 117 characters in early December to the high 80s by February and March, then to about 53 characters in April. Phrasing that matched last quarter's queries can go stale.
Other shifts are sharper. On August 8, 2026, ChatGPT started using the site: operator at scale, jumping from about 0.4% to about 17% of fanout queries. On August 14, reddit.com's share of ChatGPT citations fell from roughly 4% to 0.5%.
A strong agency re-audits after changes like these and tells you about them. Daily tracking is a reasonable baseline; ask if faster cadences are available for launches.
4. Do you check whether AI crawlers can actually read my site?
Visibility problems are often access problems. Ask whether the agency reviews robots.txt, CDN rules, rendering, and server logs for AI bots, and whether it can show you which pages GPTBot, ClaudeBot, or PerplexityBot fetched and which returned errors. Crawler mix also moves: Meta-WebIndexer grew from about 2% to nearly 38% of tracked AI crawler requests between mid-July and August 9, 2026.
Red flag: the agency has never looked at AI crawler logs. Then it cannot tell an access problem from a content problem.
Part 2: Reporting questions
5. Do you report citations, mentions, and recommendations separately?
They are different things. A citation is a link to your page. A mention is your brand named in the text. A recommendation is the AI telling the user to pick you. A brand can be mentioned often and never cited, which usually points at content or crawlability. The reverse is a prominence problem. Promptwatch's citations vs mentions guide explains the distinction.
6. Which metrics will be in the monthly report, and how are they defined?
Look for visibility score, share of voice, average position, sentiment, and citation share by topic, each with a written definition. Ask for a sample report from a current client with the name redacted.
Red flag: the report is rankings, impressions, and domain rating with an "AI" slide at the end. ScaleTheory's guide lists exactly this as the vanity-metrics pattern.
7. How do you separate platform changes from the effect of your work?
This is the question I'd weight most heavily. When ChatGPT's average citations per response dropped around the GPT-5.3 rollout on March 4, 2026, every brand's numbers moved. An agency that claims credit for gains and blames the engine for losses is not measuring anything.
A strong answer mentions a control: a set of prompts or competitors that should not move because of your changes, used to isolate what is yours.
8. Can you tie AI visibility to traffic, leads, and revenue?
Attribution here is imperfect, and honest agencies say so. They should still describe the method: AI referral traffic from analytics, conversion events, CRM stages, and what is documented influence versus claimed revenue. Be aware that ChatGPT stopped sending a referrer on October 8, 2026, so tagging links with utm_source=chatgpt.com matters. Ask how the agency handles that.
Percepture's SEO hiring guide frames the red flag well: every organic lead converted into revenue without an approved method.
Part 3: Results and contract questions
9. Can you show a measured before and after, with method and dates?
Good case studies name the prompt set, the engines, the start and end dates, and the result: citation share, AI referral traffic, or pipeline. ScaleTheory's own example, Flowcart going from a new domain to 200+ monthly LLM sessions in four months, is the shape you want, even though it is a vendor claim you should verify. Ask for a reference you can call.
10. What will you change on my site, and what happens when a number drops?
Listen for named on-page work: content gaps filled, pages refreshed, schema and entity cleanup, internal linking, third-party mentions. Then ask about the response to a drop. Promptwatch's content half-life data tracked pages ChatGPT cited on July 6, 2026 for 90 days: news lost half its pages within 36 days, while landing pages and docs lasted longest. So decay is normal, and the plan should include refreshes, not only new content.
Content type matters too. In Promptwatch's data for Sep 28 to Oct 4, 2026, product pages were the most cited type in ChatGPT at 22.4%. An agency that only writes blog posts may be aiming at the wrong target.
11. Is earned media in scope, and how do you build third-party authority?
AI engines pull from sources beyond your site. The mix shifts constantly: in the Sep 28 to Oct 4 data, ChatGPT drew 0.46% of citations from social platforms while Google AI Overviews drew 12.2%. Ask which off-site sources the agency targets, per engine, and why. Be skeptical of anyone promising "secret tricks" or bulk link buying.
12. What do you not guarantee, and what do I keep if we part ways?
Strong answer: no guaranteed citations, a pilot of 30 to 60 days, a month-by-month plan for the first three to six months, and written confirmation that you own the pages, prompt list, and raw data. A 12-month retainer pushed up front is a red flag. Published agency starting retainers in the sources I reviewed ranged from roughly $10K to $25K a month, all self-reported, so compare scope rather than price alone.
Scoring your shortlist
Turn the answers into numbers. This weights the areas where agencies differ most.
| Area | Questions | Weight | Strong answer looks like |
|---|---|---|---|
| Tracking method | 1 to 4 | 30% | Shared prompt list, per-engine real-UI data, regular retests, crawler log review |
| Reporting | 5 to 8 | 30% | Separate citation, mention, and recommendation metrics, controls, defined attribution |
| Proof | 9 to 10 | 25% | Dated before and after with method, named on-page changes, refresh plan |
| Contract and ownership | 11 to 12 | 15% | Pilot, no guarantees, you keep the data |
Score each question 0 to 3 and keep the written answers next to the score. Discovered Labs' checklist puts reporting and attribution at 25% of a decision matrix; I'd go slightly higher, since everything else depends on it.
Red flags at a glance
| You hear | What it usually means |
|---|---|
| "We guarantee you'll rank in ChatGPT" | No understanding of how variable AI answers are |
| "Our proprietary score" with no definition | Unauditable reporting |
| One blended AI number | Hides engine differences |
| Report is rankings and traffic only | SEO shop with a new label |
| Annual contract before any pilot | Pressure instead of proof |
| No access to raw data | Lock-in |
What tooling should the agency use?
Ask which platform they use and whether you can see it directly. Agencies commonly use Profound, AthenaHQ, Otterly, Peec AI, or custom scripts. The better arrangement is read access to the same dashboard the agency works from, so you are not relying on screenshots.
For the tracking and reporting layer, Promptwatch is an end-to-end option: it monitors real-interface answers across ChatGPT, Gemini, Perplexity, Claude, AI Overviews, and others, logs IP-verified AI crawler visits, and measures AI traffic on your site. It also writes and publishes content against the gaps it finds, so an agency using it can show both the problem and the fix. Agency plans start at $199/mo; confirm current pricing at promptwatch.com/pricing. 1001 SEO Media, the agency behind this site, uses it for its own GEO work.

If you want to compare monitoring-only options your agency might propose, these are common ones:
Profound

Otterly.AI

The GEO software directory at bestgeosoftware.com lists more. Tracking-only tools answer "was I mentioned?" The harder question, what to do about it and whether it worked, is where agency labor should show up.
A quick call script
If you only have thirty minutes with each agency, use this order:
- Ask for the prompt list and engine coverage (questions 1 and 2).
- Ask them to walk through a real report and point to where citations and mentions are split (5 and 6).
- Ask how they handled the last major engine change for a client (3 and 7).
- Ask for one dated case study and a reference (9).
- Ask what you own when you leave (12).
Agencies that can answer all five without hand-waving are rare. That is the point of asking.
What to do next
Run the twelve questions on every agency on your shortlist, score them, and ask for a 30 to 60 day pilot with a defined prompt set and engines. At the end of the pilot, check whether the agency's own report matches what you see in your analytics and crawler logs. If the numbers agree, you've found a partner. If they don't, you found out cheaply.
For a free baseline before any call, Promptwatch's free AI visibility report shows how ChatGPT, Gemini, and Perplexity currently describe your brand. Bring it to the first meeting and ask the agency what it would change.

