Every AI search vendor now says it does more than monitor. Some write recommendations. Some generate content. Some say they "implement". The words are close enough that a buyer cannot tell them apart from a demo. This playbook is a set of questions that can.
It is written for the person who has to choose: a founder, a head of marketing, or an agency lead deciding whether to buy a tool, hire an agency, or do it in-house. The questions apply to all three, including us.
Why does the buying decision matter more than it used to?
Because buyers already ask the engines. Forrester reported in January 2026 that 94% of business buyers used AI somewhere in their buying process, up from 89% the year before. If the answer to "best [your category]" names three competitors, the buyer's shortlist was written before your site was visited. Spending a quarter on a vendor that only reports that fact is a quarter the shortlist kept going to someone else.
What are the ten questions to ask?
Ask them in a call, and ask to be shown rather than told. Each one comes with the answer you want and the answer that should worry you.
1. What exactly do you deliver in a typical week?
Want: a list of finished artifacts, such as rewritten passages, new articles, pull requests, directory listings and pitch drafts, with examples from a real account. Worry: "insights", "recommendations", "strategy", or a content brief someone on your team still has to write.
2. Where does the work land?
Want: in your own systems. A pull request in your repository, a draft in your CMS, a draft in your mailbox. Worry: a portal you have to copy from, or a hosted page on the vendor's domain.
3. Who approves before anything goes live?
Want: you, by default, with a clear list of what can ever go out without asking. Worry: "we handle it" with no setting you can see. In Attensira, new workspaces start in Draft: nothing leaves the workspace until you approve it, and claims such as prices, numbers, product names and competitor names wait for a person in every mode.
4. What happens if I never press approve?
Want: nothing goes out, and the vendor tells you what is waiting. Worry: work that publishes after a timeout. Ask the question literally; the answer reveals how the system is built.
5. How do you decide what to work on first?
Want: the questions you lose, ranked, each with the pages that decided the answer and whether you are on them. Worry: a generic checklist applied to every customer in the same order.
6. Do you work off my site as well as on it?
Want: yes, with rules. Most of what decides an AI answer is somebody else's page, so a vendor that only edits your site leaves the biggest lever alone. Ask who sends the pitches (it should be you, from your mailbox) and whether community replies are drafted for you to post or posted under someone else's name.
7. How do you prove a change worked?
Want: the same tracked questions read before and after, over windows of days, with the sample size, and "no change" reported when it is true. Worry: screenshots of single answers, or a blended score that only goes up.
8. What will you not claim?
Want: no guaranteed rankings, no guaranteed share of answers, no revenue attributed to a change without evidence. A vendor that can tell you what it cannot promise is one you can trust with the rest.
9. What do I pay when you deliver less?
Want: you pay for work delivered, against a budget you set, and the gap comes back to you. Worry: a flat retainer with no list of what shipped. Attensira does not charge for work it does not deliver; the shortfall is credited back on the 1st.
10. Can I read everything without asking you?
Want: every answer, every number with its sample size, every piece of work and every charge, readable in the product and over an API. Worry: a monthly PDF as the only window.
What are the red flags?
- A demo that never leaves the dashboard. If the vendor cannot open a real pull request, draft or listing from a real account, the product stops at the chart.
- Numbers without a sample size. "Visibility up 40 points" means nothing without how many answers it came from.
- Guaranteed outcomes. No one controls what an engine writes.
- Publishing on your behalf by default. Speed is not worth a wrong price on your homepage.
- Outreach from the vendor's domain. Editors answer people, and your sender reputation is yours to spend.
What should week one produce?
Hold any vendor to a concrete first week:
- A baseline. Your tracked questions read on each engine, with the answer text stored and every share shown with its sample size.
- The ranked losses. The questions you lose, and for each one, the pages that decided it.
- The first finished work in your approval queue. Not a plan for work, the work itself.
- A date for the first verdict. When the first shipped change will be read again, and what "no change" would look like.
If week one ends with a slide deck, you bought monitoring with extra steps.
We hold our own site to the same bar. On 4 October 2026 the Inbox of our attensira.com workspace held 28 items waiting on a person, ten of them finished work (four new articles and six passage rewrites), and no pull request, because we had not connected our GitHub repository. Points one to three were met; point four was waiting on our own yes.
