Short answer: Start with what they promise. Rankings, a "citation rate" or results within a set number of days are promises nobody can keep, and both Google's and Bing's documentation say so. Then look at the methods: unlabeled paid articles, mass posting and bought links put your domain and brand at risk. Finally, look at the measurement: you should be able to re-run it yourself. The contract should spell out scope, deliverables, how results are measured, what is tracked but not promised, how you exit and get refunds, and who owns the pages and accounts.

Claims that should make you pause

  • Promised rankings, or "we'll make ChatGPT recommend you first". Google's own page on hiring an SEO says no one can guarantee a top ranking on Google, and warns about anyone who promises rankings, claims a "special relationship" with Google or offers "priority submit". Bing's webmaster guidelines put it plainly too: SEO does not guarantee rankings or traffic, and GEO does not guarantee grounding or citations in AI answers.
  • A "citation rate" with no method behind it. Ask the same question twice and the sources an assistant cites can change. OpenAI itself warns that ChatGPT's search results and citations can be incomplete or outdated. If a provider can't tell you which questions, which assistants, how many runs and under what conditions, you can't check the number.
  • Results promised to the day. Fixing a blocked crawler takes days. A page that answers one specific buyer question usually takes weeks before you can tell whether AI cites it. Broad category terms take months. If someone names a day count, ask what it's based on and across how many stores.
  • Unlabeled paid articles. Placing your content on news or industry sites without an "ad" or "sponsored" label. The FTC expects advertising to be recognizable as advertising, and Google's site reputation policy covers third-party content published mainly to borrow a host site's ranking signals.
  • Mass posting and mass pages. Hundreds of articles a day, the same talking points across many accounts, or hundreds of near-identical keyword pages. Google's spam policies count attempts to manipulate generative AI responses in Search as spam, and list scaled content abuse separately. For what honest community work looks like, see our Reddit and YouTube guide.
  • Buying or trading links. Google's link spam policy covers paying, or giving goods or free product, for links or for posts that contain links. Paid links are fine only with rel="sponsored" or rel="nofollow".
  • Vague methods, or no way for you to check. "We have an inside channel." "We know how the platforms rank brands." You can't verify claims like these, and Google lists a claimed "special relationship" among the warning signs.

This is not only about ethics. Google is direct about it: you are responsible for the actions of any company you hire, and if a provider creates deceptive content on your behalf, your site could be removed from Google's index entirely.

Questions to ask in the first call

About half of these come from Google's own list of questions to ask an SEO. The rest are specific to AI answers.

  1. What exactly will you do, and what does each piece of work deliver? Can I see past deliverables?
  2. Do you follow Google Search Essentials? Is your advice on AI answers in line with Google's guide to generative AI features? Do you cite official documentation for your recommendations?
  3. How do you measure results? On which assistants, with which questions, how often? Can I re-run it myself?
  4. What do you expect to see, and when? What won't you promise?
  5. What will you change on my site, and will you tell me about every change and why?
  6. Whose identity is used for posts and articles off my site? Are paid placements labeled as ads?
  7. Have you worked with stores in my category, selling into my country?

One more from Google: if they offer an audit first, give read access to Search Console only. Hold back write access at that stage.

Six clauses to put in the contract

ClauseWhat it coversWhat clear wording looks like (illustrative)
ScopeWhat they do, and what they explicitly don't"Fix crawler access, write question-answer pages, re-measure monthly. No category-term ranking work, no off-site article placement."
DeliverablesWhat arrives each month or quarter, and whenChange log, published pages with a fact-check sheet for each, monthly report, each with a due date
MeasurementA score you can re-run, plus a fixed question setWhich rubric and version, the baseline date; the question list attached as an appendix and frozen for the term
Tracked, not promisedNumbers you watch together but don't sign off onShare of AI answers that mention you, AI referral visits, leads and orders: in the monthly report, not in acceptance
Exit and refundsNotice period and what happens to money paidWritten notice period; whether the current period is refundable; how an annual prepayment is prorated; any first-period refund
OwnershipWho owns pages, copy, images and accountsContent is yours. Search Console, Analytics, your store admin and community accounts stay in your name; the provider gets collaborator access, removed when the work ends

"Tracked, not promised" is the clause people skip. Traffic and leads matter, but at the start there is no attribution baseline, and both move with seasons, ad spend and pricing. If they rise, nobody can say how much the provider did. If they don't, nobody can say whose fault it is. Put them in the acceptance terms and you will be arguing on sign-off day.

The measurement clause is the one ledger both sides can agree on. If the rubric is public and you can run it yourself, the score isn't just the provider's word.

"Pay for performance" and "full refund if it doesn't work"

Some providers sell on performance pricing or a money-back promise. First ask what "performance" means.

  • If it's traffic: traffic moves with seasons, ads and pricing, whatever the provider does. The fastest way to hit a traffic number is often something the provider controls and you carry the risk for, like bought links or inflated visits.
  • If it's a citation rate: go back to the questions above. Which questions, which assistants, how many runs, what conditions, and can you re-run it?

A generous refund policy doesn't replace a measurement you can check.

How we do it

Briefly, so you can hold us to the same standard:

  • We don't promise rankings, placements or traffic. Acceptance is the AIV score's baseline and lift: geo-score's public rubric, 0 to 100, open source, re-runnable by you. Whether AI mentions you and how many visits it sends goes into your reports (monthly on the quarterly service, weekly on managed), outside acceptance.
  • We bill by the quarter. Prices, what each plan includes, and how exits and refunds work are on the pricing page.
  • No bought links, no keyword-swap pages, no sock puppets, no unlabeled paid articles. Each rule and its basis is on what we refuse to do.

Common questions

A provider showed me a screenshot of ChatGPT recommending us. Does that count?

It counts as one sample. Answers vary from run to run and with the account, chat memory and location. What you want is a log of a fixed question set asked repeatedly, week after week. Our weekly check guide shows how.

They say they have an inside line to the AI platforms. Should I believe it?

If you can't check it, don't count on it. Google's page on hiring an SEO singles out claims of a "special relationship" as a warning sign. A method worth paying for is one you can understand and re-measure yourself.

Should I give a provider admin access?

Read-only during an audit. Once you hire them, keep every account in your name and grant collaborator access, and remove it the day the engagement ends. Ask for a list of every change they made.

Want to know where your store gets stuck? Test it yourself with geo-score, or send us your store's URL and get a free quick review within 48 hours.

Sources

  1. Do You Need an SEO? (Google Search Central, last updated 2026-06-05), checked 2026-10-08
  2. Google's Guide to Optimizing for Generative AI Features on Google Search (Google Search Central, last updated 2026-07-10), checked 2026-10-08
  3. Spam policies for Google web search (Google Search Central, last updated 2026-08-28), checked 2026-10-08
  4. Bing Webmaster Guidelines, checked 2026-10-08
  5. Searching the web with ChatGPT (OpenAI Help Center), checked 2026-10-08
  6. Native Advertising: A Guide for Businesses (Federal Trade Commission), checked 2026-10-08
  7. geo-score README (open rubric, re-runnable score), checked 2026-10-08