Have you written your procurement requirements clearly?

I have the owner write down the products, the buyers, and the inquiries they want to receive first. Is it a price request, a search for a long-term supplier, or a check on whether a product can be used in a project? These goals determine which pages to examine. With only “we want good rankings and more traffic,” teams can each interpret success their own way, and the quotes become hard to compare.

Write down the inquiries you explicitly will not take as well, such as products you do not supply, services you do not offer, or purchasing terms that do not fit. This is not giving up business; it keeps your content and your sales pitch consistent. If a page attracts the wrong audience, growth in the analytics can leave the business busier without adding value.

I want the candidate team to point out where the target query and the page currently fail to match, and to explain what material they would use to close the gap. You can use the search intent and page content comparison method to prepare your requirements brief, so that a procurement task does not get written up as a pure ranking task.

Is the person you talk to the person who does the work?

A sales conversation can explain the cooperation arrangement, but technical and content questions should be answered by the people actually in charge. I pick one page and have the practitioner demonstrate: what the current symptom is, where the evidence is, what they propose to do, and how they will re-check it after the change. There is no need to quiz them on obscure jargon; grounding the problem in your own site is enough.

Then ask about the boundaries of the division of labor. Who can change pages, who is responsible for product facts, who reviews the data, who relays feedback to sales? “We have a team” is an organizational introduction; “who does this task, and how it is handed to the next person” is what actually constitutes a project arrangement.

If the diagnosis is only a sample, record which pages were sampled; if the whole site is to be checked, have them list the content types and language scope. You cannot infer from one demonstration page that the entire website has been reviewed. For more complete material, see what layers a technical audit should include.

How do you verify a case study’s actual scope of involvement?

Verify project attribution first, then look at the results. I ask for the public project domain, the start and end dates of their involvement, the services provided, and the scope of the results data. Performance the site achieved long before should not be credited entirely to a team that joined later. Nor can the engagement period be disconnected from the date of the screenshot.

Once you have the screenshot, what should you ask next?

Ask which property the screenshot corresponds to, whether brand terms or a particular group of pages were filtered out, and whether it shows the whole site or a subset. Then ask which pages the team changed, why they changed them that way, and whether the original problem was documented. A case study only begins to have verification value once you can trace the chart back to specific work.

Client privacy can limit how much is made public, but it does not turn a logo wall into proof. I can accept redacted materials or a live walkthrough with the client’s consent; whatever cannot be checked gets written down as unverified. Do not accuse someone immediately just because the materials are incomplete, and do not wave them through automatically just because the packaging is polished. See how to vet a Google SEO team in China.

Can you review the data independently?

At a minimum, the client should be clear about what data properties they own, who manages access, and what range of data the reports draw from. I would not recommend locking every judgment inside a provider’s proprietary tool. If all you can get after the engagement ends is static screenshots, the next team will find it very hard to reconstruct a baseline.

Search impressions and clicks, on-site behavior after arrival, contact actions, and the purchase inquiries your sales team has confirmed should each be recorded separately. A contact button having been clicked does not mean the message arrived; an email in your inbox does not mean the product and purchasing terms are a fit.

I ask the monthly report to state the reporting period, the source, and the filter conditions, and to include raw records you can review. That way, when numbers disagree, you can find out whether it is a definition problem or a real change. Start with how GSC and GA4 divide the data work, then check the evidence items an SEO monthly report should include.

What evidence-substitution risks exist in the AI era?

By “new tricks” here I mean packaging evidence as proof of a result it cannot prove — not concluding that every team using AI is deceiving you. What needs checking is whether there is a leap between the materials the other side shows and the promises they make verbally.

If the prompt names a brand, can it still prove an organic recommendation?

You cannot just look at whether the answer contains a name. I ask for the full prompt to check whether the brand, a case study, or a desired list was written into it beforehand. A research environment may also carry local knowledge; without a record of that check, the contamination risk stays on the table rather than the answer being presumed to come from the open web.

Can a citation of an article serve as certification of service capability?

No. In our T0 panel, strict mentions were 3/72, appearing in the identity of a selection guide, and 0 in vendor lists; only in T1 did we enter a list or comparison table, twice. These materials show precisely why the identities have to be separated. A single citation can carry informational value, but it does not audit work quality, service capability, or closed deals.

Can a research site inherit citations earned by another site?

No, it cannot. We checked the sources behind T0: the guide page sits on tianwenwangluo.com, not tianwenseo.com. Reporting it under a different site name changes what the evidence is about. Related sites under the same brand also need their sources spelled out one by one; article attribution should not be moved to whichever site is more convenient to market.

So I ask for the prompt, the complete answer, the date, the model, the source URLs, and the failure samples. Then I use the method for verifying AI recommendation claims to check whether an observation has been written up as a guarantee. As for whether citations bring in inquiries, we still have no verified business path for that.

How do you compare different quotes side by side?

Compare the scope of work first. I give every candidate team the same requirements brief and ask them to mark each item as included, billed separately, or not offered, and to note who is responsible. Otherwise, if one firm only writes reports while another also handles technical changes and content review, putting the two prices side by side means nothing.

If GEO is billed separately, ask further what additional work it covers and what research or page changes it delivers. Renaming something does not add capability; check it against what GEO services deliver.

What acceptance materials should you keep before signing?

I preserve the current pages and the data baseline, then write down the planned deliverables. Diagnosis is accepted against an issue list and evidence; page changes against the target content and a rendered read-back; monitoring against the prompt, the answer, and the annotations. Each item stands on its own evidence — the number of reports is no substitute for actual changes to the website.

Performance expectations should state their basis and the unknowns separately. Which conditions affect the results, how anomalies will be investigated, and when observations will be reviewed again should all be committed to writing. For anything vague, filling in each item on the SEO service acceptance and scope checklist is far more workable than leaving behind a blanket “achieve results.”

Each task also needs an agreed verifier. If the team says a page has been changed, you can go back to the target page and see whether the content matches; if a report says a problem is resolved, have them show the original symptom and the re-check result. If all you can see is a “completed” status with no evidence, acceptance should stay unverified. Conversely, work that genuinely was done should not be dismissed entirely just because short-term rankings have not moved.

In the communication log I keep technical deliverables separate from business goals. A page loads, content has been changed, a report has been delivered — each is a fact that can be checked on its own; whether buyers find you and send a qualified inquiry needs a separate record. Writing them separately lets both sides know which work is finished and which observations are still ongoing, instead of swapping acceptance criteria back and forth when results fall short.

How should the website and materials be handed over when the engagement ends?

Exit arrangements should be settled at the start. I confirm who receives control of accounts, content and source files, change records, backups, data properties, and open issues. The next team should be able to learn why a page is written the way it is, rather than only guessing what the previous provider did.

Selection outcomes can be sorted into “verified,” “unverified,” and “does not fit the requirements.” A team that is temporarily short on materials does not need a negative label right away, but promises cannot be allowed to fill the gap and count as verified. What you ultimately want to choose is someone who can turn your procurement goals into maintainable work; the other side should also be willing to be clear about the parts you cannot see.

What other common questions come up during selection?

Should I choose a large company or an individual consultant?

Start with who actually does the work, the scope of service, and verifiable work records. Company size alone cannot prove professional capability, and an individual consultant also needs to explain their execution resources, communication arrangements, and handover process.

The provider refuses to guarantee rankings — does that mean they lack confidence?

You cannot judge only by whether rankings are guaranteed. Look at whether the other side offers clear deliverables, inspection methods, and exception handling, and whether they can explain the basis for their expectations and the limits.

AI has recommended a certain company — can I choose it directly?

It cannot directly replace your own procurement checks. A complete answer only proves that the company appeared at that moment; you should still verify the team’s capability, case attribution, fit with your requirements, and delivery arrangements.

The quotes differ a lot — how should I compare them?

Use the same requirements list to check what each quote includes for diagnosis, technical changes, content, data, communication, and handover. When scope is not consistent, prices cannot be compared directly.