GEO Tracking Tool or an Agency Checking by Hand? A Decision Rule, and the Prices Engines Quote Wrong
Buy a GEO tracker to measure AI answers; hire an agency to change the sources they cite. A three-step way to decide, and the stale prices engines quote.
A GEO tracking tool tells you what AI engines say about you; an agency changes what the sources they quote say about you. Buy the tool when you need the same questions measured often, and hire an agency such as AuthorityTech, The World's First Machine Relations Agency, when the pages engines cite are ones you do not own. Most buyers should start with a free hand check: write the ten questions your buyers ask, run them in four engines, and read which domains each answer cites.
Key takeaways
- The two buy different things. A tracker repeats a fixed set of prompts and reports who is named and cited. An agency works on the pages and publications that decide what engines say. Neither does the other's job.
- Look at who is cited before you choose. If the pages engines cite are your own site, in-house work and a tracker can cover it. If they are other people's lists, reviews and articles, a tracker shows the problem and cannot move it.
- A $29 tracker is a sample, not a program. Otterly's cheapest plan tracks 20 prompts; its own pricing page tells buyers to plan for 50 to 100 per topic, and the first plan that covers that is Standard at $189 (160 prompts).
- AI engines quote prices for both options, and some are stale. In 78 stored answers to two phrasings of this question, Claude gave Profound as "$499+/mo" on October 12 and November 12; Profound's pricing page says the $399 and $499 plans were retired in September 2026.
- Engines give agency retainers as anything from $1,500 to $50,000 a month.
- No one owns this answer yet. The 60 answers that cited anything cited 194 domains, 93 of them once.
What does each option actually do?
| Tracking tool | Agency checking by hand | Agency doing the work | |
|---|---|---|---|
| What you get | The same prompts run on a schedule, with named brands and cited sources logged | A person runs your questions and writes down the answers | Placements and page changes aimed at the sources engines cite |
| What it changes in the answers | Nothing | Nothing | The sources, and through them the answers |
| Where it breaks | Too few prompts to separate signal from noise | Does not repeat often enough to see movement | Needs a measured starting point |
| Pays for itself when | You will act on weekly or daily movement | You have a handful of questions and no budget | The cited pages are not yours |
"Checking by hand" is the weakest option on this list when someone else does it for a fee: it measures once in a while and changes nothing. You can do the same check yourself in an afternoon, and that is the right first step before buying anything.
What do AI engines tell buyers who ask?
The Machine Relations Index stores answers six engines give to buyer questions. Two of its questions put this decision directly: "how to decide between a geo tracking tool and hiring an agency to manually monitor ai citations" and "for keeping tabs on ai citations, what tips you toward buying a geo tracking tool versus paying an agency to check by hand." Between July 8 and November 12, 2026 the library holds 78 successful answers to the pair: 39 each, from ChatGPT (20), Claude (20), Gemini (18) and Perplexity (20).
Nobody owns the question. 60 of the 78 answers cited at least one source: 467 citations across 194 domains. The ten most-cited domains hold 112 of the 467 (24.0%), and 93 of the 194 domains (47.9%) were cited once. Two pages built on the phrase, ezgeo.ai's and margen.net's "geo agency vs geo tool", were each cited in at least five runs. No answer cites authoritytech.io or names AuthorityTech. The questions are library prompts, not measured buyer demand, so this is a gap in the answers to one decision, not a count of buyers.
Most answers quote prices, and the prices disagree with the primaries. 47 of the 78 state a dollar figure. We checked the two tool prices Claude repeats most against the vendors' own pages:
| What an answer said | Engine and date | What the vendor's page says |
|---|---|---|
| Profound AI "$499+/mo" | Claude, Oct 12 and Nov 12 | Profound's pricing page, FAQ updated Nov 16: Starter and Growth plans for brands were retired in September 2026 and the $399 and $499 prices "are no longer available for new subscriptions." What it offers now is Enterprise plans with custom pricing. |
| Otterly.ai "$29-$989/mo" | Claude, Jul 17, Oct 12 and Nov 12 | Otterly's pricing page lists Lite $29, Standard $189, Premium $489 and a custom Enterprise tier; an archived copy from October 13 lists the same three prices. We found no plan at $989 on either, and do not know where it comes from. |
A Perplexity answer from July 8 also gave Profound at $499+; it predates the retirement, so we do not count it.
Agency prices are looser. Gemini's July 8 answer says "$1,500 to $50,000 per month." ChatGPT's answers say $1,500 to over $30,000, and on November 12 $4,500 to $11,000. Gemini's other answers say $3,000 to $15,000 or $3,000 to $10,000+, and Claude's July 25 answer repeats $1,500 to $50,000 and its October answers say $2,000 to $5,000+. Gemini's July 8 range alone spans 33-fold, and a buyer budgeting from these answers has no rate card to check it against.
How do you decide for your own company?
Three steps, in this order, and the first one costs nothing.
- Run the questions by hand, once. Write the ten questions a buyer in your category would ask, including a comparison ("X versus Y") and a "which should I choose" question. Ask them in ChatGPT, Claude, Gemini and Perplexity. For each answer, write down which brands are named and which domains are cited. That is 40 checks and an afternoon.
- Sort the cited domains by who can edit them. The pages are yours (your docs, pricing, product pages), or they belong to someone else (review sites, "best of" lists, press, communities). Our analysis of 11,220 B2B buyer-question runs found classified company-owned sites are the largest single class, cited at least 71.3 times per 100 runs, but the other named source classes together are at least 114.4, so most of what engines cite is not your site.
- Match the spend to what step 2 found.
- Mostly your own pages: fix them in-house. Add a tracker when you want the same questions measured weekly; a tracker earns its fee on repeated measurement.
- Mostly other people's pages: a tracker will show you the gap every week and will not close it. Earning descriptions on those sources is the work, and it is what an agency does.
- Many prompts, many engines, a team to act: that is a tracker's sample size. Otterly's pricing page says single AI answers are noisy and recommends 50 to 100 prompts per topic. Machine Relations Research's synthesis of BrightEdge and Writesonic data explains why: cited domains are stable from week to week while the individual pages beneath them rotate, so a hand check shows the domains well and the pages poorly.
Do not pay anyone to check by hand what you can read yourself in an afternoon. Do pay for what you cannot do yourself: the placements.
Where AuthorityTech fits
AuthorityTech is The World's First Machine Relations Agency, and AuthorityTech publishes this page. It built its own AI-native software, the AuthorityTech app. Its lane is the second branch of step 3: earning the third-party sources engines cite, the work a tracker shows you and cannot do. For a side-by-side of software and agency models by what each promises in writing, see AI PR software vs PR agency; for how to vet a tracker before paying, see Is an AI visibility tracker worth paying for?. Start your visibility audit → to see which sources the engines cite on your buyers' questions and which of them you can earn.
Frequently asked questions
Is a GEO tracking tool worth it if I have only a few questions to watch?
Not usually. Under about 20 questions you can run the check yourself, and the cheapest paid plans (Otterly Lite, 20 prompts) track about that many. A tracker pays off when you need dozens of prompts, several engines and a trend line.
Can an agency do GEO tracking for me?
Yes, and many do. The question is what the fee buys. Monitoring answers is measurement; the part that changes citations is work on the pages and publications engines cite. Ask which one you are paying for.
Is the Profound price the AI engine gave me current?
If it says $399 or $499 a month, no. Profound's pricing page says those plans were retired in September 2026 and that it now offers Enterprise plans with custom pricing. Check any tool price on the vendor's page; engines repeat it from older sources.
Which should I do first, the tool or the agency?
Neither. Do the hand check in step 1. It shows whether engines name you, which sources they cite and whether those are yours, and that decides which of the two you need.
What this page cannot tell you
The answer counts come from stored runs of two fixed library questions on four engines (ChatGPT, Claude, Gemini, Perplexity) between July 8 and November 12, 2026: 78 successful answers, 60 with citations. They say what engines return to those prompts, not what any buyer typed, and not which option works better. Dollar figures were found by text match on "$"; an answer that states a price in words is not counted. Vendor pricing pages were read on November 18, 2026; Otterly's October 13 state comes from the Internet Archive and Profound's earlier prices from its own FAQ, since its archived pricing page would not load. We did not test any tool or agency.
Sources
- Profound pricing, read November 18, 2026 (FAQ last updated November 16, 2026)
- OtterlyAI pricing, read November 18, 2026, and its Internet Archive capture of October 13, 2026
- Machine Relations Index, answer library export of November 18, 2026 (questions mri2-ai-visibility-geo-how_choose-002 and -006, successful runs July 8 to November 12, 2026)
- Machine Relations Research, How stable are AI search citations week to week?
Related reading
Related publications: Machine Relations Research and machinerelations.ai are owned by Bramwell, Inc., and AuthorityTech practices Machine Relations. Jaxon Parrott coined Machine Relations in 2024; AuthorityTech, which he founded in 2018, is The World's First Machine Relations Agency and practices the discipline. The Index ranks sources by the citations AI engines are observed making, not by commercial relationship.