The First-Week GEO Audit: What We Check
A day-by-day view of our first GEO audit week, covering scope, prompts, citations, crawler access, traffic measurement, and the initial action register.
The first week of a GEO engagement is for establishing what can be measured and what is technically possible. It is too early to call a short run of answers a stable trend. We use the week to define the observation set, check access, find obvious representation problems, and produce an action register with evidence.
This order protects the client from premature content recommendations. A missing brand mention can come from a poor prompt set, an unrecognized alias, a blocked page, weak source coverage, or a genuine visibility gap. Those explanations require different owners. The audit should narrow them before anybody commissions a new article.
Before the working week begins
We confirm the engagement scope and access list. The scope names the products or services, countries, languages, competitors, and business areas under review. The access list covers the visibility platform, analytics, tag manager, sitemap or Search Console, content system, and CDN or relevant log source.
We also identify decision owners. Brand and legal teams may need to approve aliases, claims, or crawler policy. Infrastructure owns firewall and CDN rules. Content owners control publishing. Analytics or revenue operations owns conversion definitions. Putting names next to those roles prevents a technical finding from sitting in a marketing deck without an owner.
Day one: brand and market definition
We review the brand record as if we had never seen the website. That includes the official name, common aliases, domain, products, categories, locations, and short factual description. Detection can miss a mention when the monitoring system does not know a valid alias, so this is measurement setup rather than cosmetic copy.
The competitor list is limited to organizations a buyer would reasonably compare. We do not pad it with famous companies to make a chart look busy. Share of voice changes when the comparison set changes, which means the list needs a rationale and a date.
We then document market boundaries. A model may answer the same question differently by location or language. If those differences matter commercially, we keep the monitors separate instead of blending them into one score.
Promptwatch's agency GEO guide frames benchmarking as the start of an agency engagement. We use it as a platform workflow reference. The client's scope and our audit record decide which benchmark is appropriate.
Day two: prompt and answer audit
We build a first prompt set around actual decisions. Groups may cover category discovery, use cases, comparisons, direct brand understanding, and reputation where those questions fit the business. We source wording from client materials such as sales questions, internal search, support themes, and existing research. We do not invent customer demand to fill a monitor.
Each prompt receives a topic, intent note, target market, and reason for inclusion. We remove prompts that are too vague to produce an actionable answer. Direct brand prompts stay separate from non-branded discovery because they answer different questions.
Once initial responses arrive, we inspect them manually. We record whether the brand appears, its position and description, the competitors named, the sources cited, and any factual error that can be checked against an approved source. We do not treat one unusual answer as a trend.
The Promptwatch metrics overview separates mentions, citations, visibility, share of voice, sentiment, citation rank, and self-citation rate. When we use those product metrics, the audit links to their definitions and states the filters applied.
Day three: citations and page coverage
We classify citation sources by ownership and type. A client's page, an independent publication, a retailer listing, a Reddit discussion, and a YouTube video each suggest different work. We look for sources that recur across prompts and models, then open them rather than relying on domain totals.
On the client's site, we map priority questions to existing pages. The audit asks whether a suitable page exists, whether it contains approved and current information, and whether internal links and page inventory make it discoverable. A content gap is not always a request for a new URL. An existing page may need a clearer answer or an outdated page may need consolidation.
We also flag inaccurate offsite descriptions, but we do not promise that a publisher or community will change them. The action register distinguishes owned edits, outreach opportunities, and observations the client cannot directly control.
Day four: crawlability and measurement
Technical review starts with bot purpose. Training crawlers, search index crawlers, and live citation fetchers are not interchangeable. We inspect robots.txt, CDN and firewall rules, response codes, redirects, and returned HTML for the priority URL set. Every allow or block recommendation is checked against each provider's current official documentation.
Promptwatch's crawlability guide explains the categories and common failure points visible in crawler logs. We compare that vendor guidance with the client's actual edge evidence. An absent crawler is recorded as absent during the observed period, not automatically labeled blocked.
Traffic instrumentation receives its own test. We confirm whether identified AI referrer visits are being collected, whether internal visits can be excluded, and whether landing pages line up with the main analytics system. Referrer data is a floor because some visits have no referrer and some answer exposure may lead to a later direct or branded visit.
Conversion and revenue stay in the client's analytics or CRM. The audit records the event definition and proposed join, but it does not claim revenue from visibility data alone.
Day five: findings and the next observation period
The final working session sorts findings by evidence, impact on measurement, and owner. Setup defects come first: wrong aliases, irrelevant prompts, mixed markets, broken tracking, or inaccessible pages. Content and offsite opportunities follow only when the observation supports them.
The deliverable contains the measurement scope, prompt inventory, initial response examples, citation map, crawler-access matrix, analytics status, and action register. Every action has an owner, dependency, and verification method. We mark the initial charts as a snapshot, not a month-over-month baseline.
Promptwatch's 90-day program advises using the first month to establish a baseline before diagnosing movement. We follow that restraint after the audit week. Setup corrections happen promptly, while performance conclusions wait for a more representative observation window.
We run client audits on Promptwatch because monitored answers, citations, crawler records, and identified referral visits can be reviewed in the same project. At the end of week one, we expect a cleaner measurement system and an assigned queue. We do not promise a visibility lift or present seven days of collection as proof that an optimization worked.