← Insights
Choosing an agency

Can a GEO agency guarantee results in ChatGPT?

No. No agency controls how ChatGPT picks its sources: OpenAI states that placement in ChatGPT search is not guaranteed, and the same question returns different brands from one run to the next. An honest agency commits to the means it controls. ZivRank commits to at least 50 measured questions, content its team publishes, a ±15% noise threshold set in advance and objectives reported even when missed.

·6 min read·By Jonathan Taieb, Founder and CEO
Key takeaways
  • OpenAI says placement in ChatGPT search is not guaranteed.
  • In the US, the FTC rule on consumer reviews and testimonials bans fake reviews, one of the shortcuts behind some “guaranteed” offers.
  • AI referral traffic was 0.32% of visits in SE Ranking’s 2026 sample, so count named answers, not only clicks.
  • In our published US case, two objectives were missed, including ChatGPT, which stayed at 6.8% of answers against a 10% target.

Why can’t a GEO agency guarantee a place in ChatGPT’s answers?

Because no agency controls the systems that build the answer. ChatGPT chooses its searches, its search providers and its ranking factors, and OpenAI states plainly that placement is not guaranteed.

OpenAI’s help page describes ChatGPT search: it rewrites a question into one or more targeted searches, retrieves results from third-party search providers and partners, and ranks them on several factors [1]. Each step sits outside an agency’s reach:

  • Variation between runs. In a SparkToro study, ChatGPT and Google’s AI almost never returned the same list of brands twice across 100 runs [2].
  • Search providers. Which pages an index returns for a rewritten query is decided by that index, not by your agency.
  • Model and product changes. AI vendors update their models and search features, and answers can change with them.
  • Competitors. They publish too, and a source that named you this month can name them next month.

What can a GEO agency legitimately commit to?

Everything it controls: the questions measured, the protocol, the content produced and published, the delivery dates, the data handed over and the honesty of the report. Those commitments can be written into a contract and checked.

CommitmentCan it be guaranteed?How you check it
Buying questions measured, agreed before the baselineYesThe list attached to the contract
Runs per question and engines coveredYesThe raw answers
Noise threshold set before any comparisonYesIts value, date and method in writing
Content produced and published each monthYesLive URLs and publication dates
Re-measurement with the same protocolYesSame questions, engines and runs in both reports
A mention or a position in ChatGPTNoMeasured and reported, never promised
Leads or revenue from AI answersNoTracked on your own numbers

What do “guaranteed results” offers usually hide?

A guarantee is usually kept by changing what is measured or by taking risks with your brand. Before you sign one, read exactly which number is guaranteed, on which questions, and how it will be counted.

  • Branded questions. Questions that contain your name are answered with your name; counting them inflates any score.
  • A single run, or a screenshot picked among many.
  • A moving baseline, redefined after the first measurement.
  • Risky tactics: fake reviews, which the FTC rule bans [3]; vote or account manipulation, which Reddit’s rules prohibit [4]; scaled pages with no value, which Google’s spam policies target [5].

A guarantee also shifts the provider’s incentive from your buyers’ questions to whatever number the contract names. The risks of the shortcuts land on your brand, not on the agency.

Who controls the sources AI engines cite?

Mostly third parties. On the US buyer questions we measured, the domains cited most often were community, video, reference and professional platforms that neither a brand nor its agency owns.

Our November 8, 2026 measurement asked ChatGPT, Claude, Gemini and Perplexity 65 US buyer questions about GEO, 5 runs each (1,300 answers + 325 Google searches) [8]. reddit.com was cited on 59 of 65 questions (251 citations), youtube.com on 53 of 65 (235 citations), en.wikipedia.org on 36 of 65, linkedin.com on 34 of 65 and searchengineland.com on 26 of 65. The answers named 57 distinct companies, so most questions are contested. An agency can contribute to these sources within their rules, through disclosed participation, useful videos and facts sent to editors. It cannot decide what a moderator keeps, what an editor writes or which page the engine retrieves.

What ROI can I expect from generative engine optimization, and how fast?

Expect visibility changes after a few weeks and pipeline effects later. The ROI is calculated on your own numbers, against a baseline measured before any work; without a baseline, no return can be shown.

Across ZivRank programs, new content has been picked up in answers 3 to 8 weeks after publication. Clicks understate the effect: AI referral traffic was 0.32% of visits in SE Ranking’s 2026 sample [6], and in Pew’s US panel users clicked a traditional result on 8% of visits with an AI summary, against 15% without [7]. Many buyers read the answer and come back later through search, a direct visit or a sales call. Count named answers alongside traffic, add an “AI assistant” option to your “how did you hear about us?” field, then compare leads, conversion rate and contract value with the program cost.

How should objectives appear in a GEO contract?

As measured targets on the agreed protocol, reported whether they are hit or missed, never as a guaranteed outcome. A missed objective reported plainly tells you more than a guarantee kept by redefining the count.

Write each objective as a rate on a fixed question set: for example, the share of answers naming you on ChatGPT, measured with the same questions, engines and runs as the baseline, and read against a noise threshold set in advance. Our published US case shows why the reporting matters. Over 44 days, answers naming the brand on Perplexity went from 21 to 50 out of 250, yet two objectives were missed: the ChatGPT rate stayed at 6.8% against a 10% target, and the overall citation rate reached 29% against 30%. Both misses are on the case page.

What does ZivRank commit to?

To means, in writing: 50 buying questions measured on 5 engines with 5 runs each, a noise threshold of ±15% set before any comparison, content written, fact-checked and published by our team, and the raw answers with a monthly report.

Programs run $2,300 to $10,000 per month over 6 months, long enough to publish, wait for pickup and re-measure. The $2,000 baseline audit is billed at the start and included in the program. Objectives are set on the baseline and reported with the same protocol, including the ones we miss. We do not sell a placement in ChatGPT or in any other engine. To see where you stand before any commitment, the free Quick Scan covers up to 5 of your questions, with an expert video within 48 hours. The full protocol is on our methodology page.

Frequently asked questions

Can you manipulate ChatGPT answers? What are the risks?

People try, with fake reviews, fake forum accounts, hidden text or pages produced by the thousand. The risks fall on the brand: the FTC rule bans fake reviews in the US, Reddit’s rules prohibit manipulation, and Google’s spam policies target scaled content with no value. A buyer who spots the trick stops trusting everything else you publish.

Is a pay-per-mention GEO contract a good idea?

Rarely. Payment tied to mentions pushes the provider toward the easiest counts, such as branded questions or engines where you already appear, and it needs a counting method both sides trust. Paying for defined work, with objectives measured on an agreed protocol, keeps incentives aligned with the questions that bring you buyers.

What happens if a GEO agency misses its objectives?

A good agency reports the miss with the same protocol, shows the raw answers, explains what it will change and why. Agree in advance on what follows a miss: a revised plan, credits moved to the formats that worked, or an exit at the end of the commitment. A provider that never reports a miss is not measuring carefully.

How can I check an agency’s results myself?

Ask for the raw answers behind every number, the question list and the run dates. Then rerun a sample of questions yourself, several times each, with search on. Your counts will not match exactly, since answers vary from run to run, but the direction of change should be the same.

Sources

Every figure in this article comes from the sources below, checked on November 18, 2026. Bracketed numbers in the text refer to this list.

  1. [1] OpenAI Help Center, Searching the web with ChatGPT, checked October 11, 2026.
  2. [2] SparkToro (Rand Fishkin), NEW Research: AIs are highly inconsistent when recommending brands or products, January 27, 2026.
  3. [3] Federal Trade Commission, Trade Regulation Rule on the Use of Consumer Reviews and Testimonials, posted August 21, 2024.
  4. [4] Reddit, Reddit Rules, checked November 9, 2026.
  5. [5] Google Search Central, Spam policies for Google web search, checked November 9, 2026.
  6. [6] SE Ranking (Yulia Deda), AI traffic grew 16x from 2024 to 2026, June 18, 2026.
  7. [7] Pew Research Center, Google users are less likely to click on links when an AI summary appears in the results, July 22, 2025.
  8. [8] ZivRank, US measurement: 65 buyer questions about GEO, ChatGPT, Claude, Gemini and Perplexity × 5 runs (1,300 answers) + 325 Google searches, November 8, 2026.