Check the evidence
before you sign.
Before signing with any SEO or GEO agency, verify their claimed results yourself. This checklist shows how to read comparison windows, separate branded traffic from discovery, check AI citation evidence and distinguish a measured observation from a marketing claim.
Tendou is a provider.
This guide is written by Tendou, an SEO and AI visibility agency. The examples below include Tendou's own published client records alongside the checks you should apply to any provider, including Tendou. This is not an independent awards panel or a guaranteed outcome comparison.
Does the agency
work in your market?
Check whether the agency's published results come from the country and service category you operate in. A Singapore funeral services record and a Malaysian yoga studio record do not predict results for a US plumbing business. Ask which market each example comes from and whether the agency has done the type of work you need.
Equal periods,
stated dates.
A credible Search Console comparison states two date ranges of equal length, both clearly labelled. Look for the exact start and end dates. If a provider shows only a percentage or a graph without dates, you cannot verify the claim.
Example from a published record: Tendou's HAE Studio record shows equal 23-day Search Console windows recorded on 6 September 2026 in Malaysia, with 55 clicks in the earlier window and 305 in the later. The record states that branded searches led the returned query rows and that clicks are not bookings or revenue. Read the full HAE Studio record and its limits ↗
What to check: are the two windows the same length? Do they fall before and after the work? Does the report separate branded searches from discovery terms? A jump from 55 to 305 during an engagement is an observation, not proof that the agency alone caused it.
Small bases
distort percentages.
A 400% increase from 2 clicks to 10 is still only 8 additional clicks. Ask for the absolute numbers. When a domain migration combines two domains, the total may decline even as the primary domain grows: Tendou's Four Solutions record shows foursolutions.sg going from 85 to 89 clicks, but the two domains together fell from 138 to 90 over the same 28-day windows. Read the full Four Solutions record ↗
Who is searching
the business name?
Search Console reports include both branded searches (the business name) and discovery terms (services, questions). A branded increase may reflect offline marketing, not SEO work. Ask whether the provider separates branded and non-branded queries in their reporting. Brand growth alone does not establish that SEO or GEO caused it. Google omits some query rows for privacy and other limitations, so a returned query list is not a complete breakdown of property totals. Google's query-data limitations explain the distinction.
Question, engine,
mode, date.
An AI citation claim should state the exact question asked, which engine answered (ChatGPT, Google AI, Gemini, Perplexity), whether the session was signed in or anonymous, and the date. A screenshot of one answer on one day is a single observation, not a permanent recommendation.
What to distinguish:
- Named: the answer identifies the business by name. This alone is not a citation.
- Owned-site cited: the answer links to a page the business owns. This is a source citation.
- Third-party cited: the answer links to a directory, review platform or editorial article. The business may appear in that source, but the citation is not to its own site.
- Absent: the completed answer did not name the business.
- Unknown: the check was blocked, did not complete or was not attempted.
Example: Tendou's BJJ Singapore record shows that on 11 October 2026, a ChatGPT answer named BJJ Singapore and linked its beginner page for "recommended bjj gym singapore". Perplexity also named and linked it. Gemini did not name it for the same question. These are individual dated observations from a Singapore engagement, not a guaranteed result or a complete market survey. Read the full BJJ Singapore record and its limits ↗
Named in how many
of how many?
If an agency says a client was "named in AI answers", ask: named in how many completed answers out of how many questions checked, on which engines? A business named in 12 of 28 completed ChatGPT answers for specific pest control questions is a different result from "named in AI". The denominator matters. Blocked, incomplete or unattempted checks should be reported separately, not counted as either positive or negative.
What changed
on the website?
Ask for a dated list of the actual changes: which pages were created, rewritten, redirected or technically fixed, and on what dates. A good record separates the work done from the results observed. If a provider cannot tell you what they changed and when, there is no traceable implementation record to compare with the result. Even a dated change log does not prove that one change caused later movement.
Check who owns the website, the hosting, the Search Console property and the analytics account after the engagement. These should remain yours. Giving a provider access while keeping ownership ↗
A failed check
is not a result.
If an AI platform blocked the check, required sign-in, or returned an error, the result is unknown. It should not be reported as absent (proof the business was not named) or as a success. Ask any provider how they handle incomplete checks in their reporting.
Our own records,
the same checks.
Tendou publishes dated records with the same structure this guide recommends. Each record states the market, dates, observations and relevant limits. Records with Search Console comparisons identify their comparison periods; AI records distinguish naming from an owned-site citation. The HAE Studio record says branded searches led its returned query rows. The Four Solutions record retains the combined-domain decline. The BJJ Singapore record notes that Gemini did not name the business for the same question where ChatGPT and Perplexity did.
None of these records claims a guaranteed outcome or isolates Tendou as the sole cause. They are observations during engagements. Apply the same worksheet to them that you would apply to any other provider.
What this does
not promise.
This guide helps you read agency evidence. It does not predict rankings, AI citations or enquiries for your business. AI answers change between checks. Google's ranking factors and AI features may change after this guide was published. No figure in this guide is a forecast or a guarantee.
For an outside reference, Google's guidance on hiring an SEO supports asking for prior work, clear measurement and explanations of proposed changes. Its AI Search guidance does not promise inclusion or require special AI files or schema.
Checking agency evidence.
What should equal comparison windows look like in a Search Console report?
Both periods should cover the same number of days, for example 28 days against 28 days. Check that the compared windows do not overlap and that both fall within the engagement period or clearly label which is before and which is after.
How can I tell whether an AI citation is an owned-site link or a third-party mention?
Open the citation source in the AI answer. For the business being measured, count an owned-site citation only when the answer links to that business's own website. An agency case study about a client is a third-party citation for that client; it is not a citation to the client's own site. If it links to a directory, review platform or editorial article, it is a third-party source. A named mention without any source link is not a citation.
Should I trust percentage increases from small starting numbers?
A large percentage from a small base can be misleading. Going from 2 clicks to 10 is a 400% increase but still only 8 additional clicks. Ask for the absolute numbers alongside any percentage claim.
Apply this to your own website
What's holding
your business back?
A website that needs work. Customers who can't find you. A team running on spreadsheets and messages. Start there.