Should you hire a GEO agency or do generative engine optimization in-house? For most companies the answer is a split: keep the facts about your business, the approval of anything published in your name, and ownership of the measurement in-house, and hand off the specialist execution (technical audits, measurement operations, content production and outreach) to an agency or consultant if you lack the hours or skills. A good GEO agency will be explicit about its methodology, report results per AI engine and per question, and never promise placement. Google's own guidance on hiring search help states that no one can guarantee a #1 ranking, and the same logic applies even more strongly to AI answers, which vary from run to run. Treat any guarantee of appearing in ChatGPT or AI Overviews as the clearest red flag in this market.

This is an opinion piece, written by a company that builds measurement software rather than selling agency services. We have tried to keep it neutral: what each model is good at, how to evaluate an agency or consultant, and the claims that should end a sales conversation.

Key takeaways

  • Most GEO work is not new. Google lists "optimizing for generative AI" as one of the services ordinary SEO providers offer, and says there are no special optimizations needed to appear in its AI features.
  • What is genuinely specialist is measurement: running a fixed question set across several engines, counting brands and citations correctly, and separating real change from noise.
  • Keep the truth in-house. The facts about your products, the approval of anything published under your name and the question set that defines success should never belong to a vendor.
  • Evaluate an agency on its methodology, not its case-study screenshots. Ask how it counts citations, which questions it excludes and how it handles changes in its own measurement.
  • Guarantees of placement, claims of special relationships with AI companies and fake community activity are disqualifying.

What a GEO agency actually does

A GEO agency, or generative engine optimization agency, sells help with how a brand appears in answers written by AI systems such as ChatGPT, Perplexity, Gemini, Claude and Grok. The label covers very different businesses: established SEO agencies that have added an AI service line, digital PR firms that now pitch AI visibility, content studios, technical consultancies, and a growing number of specialist shops that sell nothing else. Before comparing an agency with an in-house team, it helps to break "GEO services" into the work they actually contain.

The five workstreams inside GEO services

  • Technical access. Making sure AI crawlers can fetch and read your pages: robots rules, firewall and bot-management settings, JavaScript rendering, canonical URLs, dates.
  • Entity and facts. Making sure the models know which entity you are and describe it correctly: consistent naming, Organization markup, a clear statement of category, correction of stale facts.
  • Content. Writing pages that answer the questions buyers ask, in sentences an assistant can lift: comparison pages, use-case pages, pricing in plain text, evidence pages.
  • Third-party corroboration. Getting the same facts confirmed on the independent sites assistants cite: publishers, directories, review profiles, industry bodies.
  • Measurement. Running a fixed set of buyer questions across engines on a schedule, recording which brands are named, recommended and cited, and reporting change honestly.

The first four are extensions of work that SEO, content and PR teams already do. The fifth is the part that is genuinely new, and it is also the part where the quality of providers varies most.

Four kinds of provider, four different strengths

Because "GEO services" is a new label on old and new work alike, the type of firm behind a pitch tells you a lot about what it will actually do well.

  • SEO agencies with an AI service line are usually strongest on technical access and site content, and know how to work with your developers. Ask how their measurement differs from rank tracking; some simply rename keyword reports.
  • Digital PR firms are strongest on third-party corroboration: getting accurate coverage on the publications and directories that assistants cite. Ask how they choose targets; citation data is a better guide than audience size.
  • Content studios are strongest on producing pages at volume. Ask how they ground drafts in your facts and who checks them; fluent content with invented specifics is worse than none.
  • Specialist GEO shops are often strongest on measurement and cross-engine patterns, and weaker on implementation. Ask who will make the changes they recommend.

None of these is better in general. The right one depends on which workstream is your bottleneck, and many companies end up with two: one for measurement and planning, another for execution.

How much of it is relabelled SEO

More than many sales decks suggest. Google's guide to hiring an SEO now lists "optimizing for generative AI" alongside site reviews, technical advice, content development, keyword research and SEO training as useful services that SEOs, agencies and consultants provide. Google's documentation on its AI features is blunter still: there are no additional requirements to appear in AI Overviews or AI Mode, no special optimizations necessary, and no need to create new machine-readable files, AI text files or markup to appear in them.

That does not make GEO a fiction. Other engines retrieve and cite differently from Google, answers name brands rather than rank pages, and the research literature does find content changes that matter: the original GEO paper (Aggarwal et al., accepted to KDD 2024) reports that its methods can raise visibility by up to 40% in generative engine responses on its benchmark, with citing sources, adding quotations and adding statistics performing best, while keyword stuffing offered little to no improvement. But it does mean that a provider whose pitch rests on secret techniques, special files or proprietary "AI ranking factors" is selling something that the largest search engine says does not exist for its own products.

What is genuinely specialist

Measurement is where specialist skill pays. Doing it well means writing questions that reflect real buyer situations, running them on several engines repeatedly, resolving brand spelling variants, counting only the sources actually cited in an answer, keeping track of what changed in the measurement itself, and reporting per engine and per question. Done badly, it produces confident numbers that move for reasons that have nothing to do with your brand. Most of the evaluation advice later in this piece is about telling those two apart.

GEO agency vs in-house: the real trade-offs

The choice is rarely all-or-nothing. Most organisations end up with some version of a hybrid, and the useful question is which workstreams belong where. Before that, it is worth being clear about what each model is structurally good and bad at.

FactorIn-house teamGEO agency or consultantHybrid
Knowledge of the businessDeep: product, customers, roadmap, historyShallow at first; depends on your briefingFacts stay in-house, agency works from them
Specialist skill and toolingMust be built or boughtAvailable immediately, shared across clientsRented where needed
Speed to startSlow if hiring is neededFastFast for execution, slower for ownership
Cross-industry pattern recognitionLimited to your own categoryBroad, if the agency measures consistentlyAgency supplies it
Access to approversDirectIndirect; every change needs a sponsor insideSponsor inside owns approvals
ContinuitySurvives vendor changes; at risk when people leaveEnds with the contract unless data and assets transferDepends on who owns the data
Cost structureSalaries and tools; fixedRetainer or project fees; variableBoth, smaller of each
Main riskDoing too little, or measuring in a way that flattersPaying for activity, or for numbers you cannot verifyUnclear ownership between the two

Who uses agencies today

The best recent primary data on this comes from small and mid-sized businesses rather than enterprises. In WordStream's 2026 survey of more than 300 small businesses, half managed SEO in-house, rising to 57% of solopreneurs, while 19% of one-person brands used an agency against 53% of the largest businesses in the sample. Familiarity with GEO followed the same gradient: 65% of the largest businesses said they were extremely or very familiar with it, against 39% of mid-sized businesses and 34% of single-employee organisations.

Use an agency for SEOShare of businesses in each size groupOne-person businesses19%Largest businesses53%Extremely or very familiar with GEOShare of businesses in each size groupSingle-employee34%Mid-sized39%Largest65%Two separate questions, separate axes. WordStream by LocaliQ survey of 300+ US small businesses, 2026.
Figure 1. Larger businesses are far more likely to use an agency for search work and to be familiar with GEO. Two separate survey questions on separate axes. Source: WordStream by LocaliQ, Small Business Website Trends Report 2026, 300+ US small businesses.

The pattern is unsurprising but useful. Larger organisations have more to gain from specialist help, more budget to pay for it and more internal capacity to manage it. Very small businesses mostly do their own search work and are less familiar with GEO, which makes them more exposed to a confident pitch they cannot evaluate.

What teams find hard

The same survey asked about the biggest GEO challenges. Competitors' content appearing in AI summaries instead of their own came first at 24%, with difficulty tracking and attributing AI referral traffic and declining organic traffic tied at 22%. No single GEO challenge received more than 30% of responses, which suggests that teams are not yet sure where the problem is.

Biggest GEO challenge (share of respondents)Competitors’ content shown instead24%Tracking and attributing AI traffic22%Declining organic traffic22%Top three answers shown; no challenge exceeded 30%. WordStream by LocaliQ, 2026.
Figure 2. The top GEO challenges are close together, and two of the top three are measurement problems. Source: WordStream by LocaliQ, 2026; only the top three answers are shown.

Two of the top three challenges are measurement problems: knowing who appears instead of you, and knowing what AI traffic is worth. That is a strong hint about where outside help adds most value, and where an agency that cannot explain its measurement adds least.

When in-house is the better default

  • Your category is narrow and technical, so briefing an outsider costs more than doing the work.
  • Your facts change often (pricing, regulations, product lines), and someone close to the change has to keep them right.
  • You already have SEO and content people who can absorb the first four workstreams with some training.
  • You can buy measurement tooling and do not need someone else to run it.

When an agency or consultant is the better default

  • You need a baseline and a plan quickly, and hiring would take months.
  • You lack technical skills to diagnose crawler access, rendering or bot-management problems.
  • You need outreach to publishers and industry bodies at a volume your team cannot sustain.
  • You want an outside view of how your category behaves across engines, from someone who measures many categories the same way.

What to hand off and what to keep in-house

A simple test sorts most of the work. If a task depends on knowing the truth about your business, or on the authority to say something in your name, keep it. If a task depends mostly on specialist skill, tooling or volume, it can be handed off, provided the output comes back to someone inside for review.

What to keep, share and hand offKEEP IN-HOUSEProduct and company factsApproval to publishRaw measurement dataSHAREThe plan and prioritiesQuestion set (you approve)Content (agency drafts,you check facts)AUTOMATE OR DROPRoutine status checksReports nobody readsHAND OFFTechnical diagnosisRunning measurementOutreach researchand pitchingNeeds the truthabout yourbusinessNeeds littleinsideknowledgeLow specialist skill or volume → High specialist skill or volumeRule of thumb: whatever depends on the truth or on the right to speak for you stays with you.
Figure 3. A rule of thumb for splitting work between a GEO agency and an in-house team: anything that depends on the truth about your business, or on the right to speak for it, stays in-house. Source: this article’s framework.
WorkstreamKeep in-houseHand offWhy
Product and company factsYes: the source of truthNoA fact a vendor got wrong becomes a wrong claim with your name on it
Approval and publishingYes: every external publication approved by a named personDrafting onlyYou are responsible for what is published, whoever wrote it
Question setOwn it; agency can draftDrafting and maintenanceIt is the denominator of every number you will be shown
Measurement dataOwn the raw results and historyRunning the measurementIf the data leaves with the vendor, so does your trend line
Technical audit and fixesImplementation by your developersDiagnosis and specificationSpecialist skill, but changes touch your stack
Content productionFact review and final editResearch, drafting, structureVolume and craft can be bought; accuracy cannot
Third-party outreachRelationships you want to keepResearch, pitching, correctionsVolume work, but must be honest and disclosed

Keep: the facts and the right to publish

The core asset in GEO is a set of accurate, current, specific facts about your business, published consistently. No agency knows those facts better than you do, and every one that is wrong in your content is a liability. Keep a single owned source of truth (a fact sheet, a knowledge base, a structured document) and require that any content produced outside is checked against it. Require that nothing is published externally without approval from a named person on your side, and keep a log of what was published, where and when. If the agency posts on your behalf on community sites or publishes on your CMS, those posts should still go through that approval.

Keep: the question set and the raw data

The question set decides what "visibility" means. An agency that writes the questions, runs them, and reports the results is marking its own homework unless you own and approve the set. Approve every question, keep the set fixed unless you deliberately change it, and require that the raw results (each answer, the brands named, the sources cited, the date and engine) are exported to you, not only summarised in a dashboard. That data is what lets you compare providers, change providers, or bring the work in-house without losing your history.

Hand off: specialist diagnosis and volume work

Technical diagnosis of crawler access and rendering, content research and drafting, publisher outreach and the operational work of running a measurement programme are all reasonable to hand off. The common pattern for each is the same: the agency produces, your team approves, and the record of what was done stays with you.

Multi-brand and multi-market organisations

Groups with several brands or markets face a version of the same choice at a larger scale. The common failure is that each brand or country hires its own provider, each provider measures differently, and nobody can compare results across the group. If you operate that way, keep the methodology central even if execution is local: one set of counting rules, one versioning scheme, a question set per market written to the same standard, and raw data that lands in one place. Local agencies can then do the work that needs local knowledge, such as publisher relationships, language and regional facts, without each one inventing its own definition of visibility.

Share: the plan

The plan itself is best owned jointly. The agency brings the patterns it has seen across clients; you bring the commercial priorities and the knowledge of which questions matter most to revenue. A plan written entirely by the agency tends to optimise for what the agency can deliver; a plan written entirely in-house tends to miss what the data is saying. Our guide to a first-quarter GEO programme is a reasonable template for the joint version.

How to evaluate a GEO agency or consultant

Most agency pitches lead with screenshots of a brand appearing in an AI answer. Those prove that one answer, on one day, named one brand. They say nothing about the method, the denominator, or whether the result would repeat. Evaluate the methodology instead.

Evaluating a GEO agency: four checks in order1. MethodWhich workstreams in-house?Which subcontracted?Curious about yourbusiness and rivals?Shares every changewith its reasoning?Stop if: secretsignals, special files2. MeasurementBrand-named questionsexcluded?Citations = sources inthe answer only?Per engine, versioned?Unmeasured ≠ 0%?Stop if: one blendedscore, no denominator3. OwnershipYou own question setand raw data?Least access first?Named approval beforeanything is published?Exit plan agreed?Stop if: data livesonly in their dashboard4. Paid trialSame questions forevery candidateSame engines, same weekRaw results exportedCompare methods,not headline numbersStop if: placementis guaranteedChecks drawn from Google’s guidance on hiring search help and from measurement rules described in this article.
Figure 4. Four checks for choosing a GEO agency or consultant, in the order to apply them, with the answer that should stop the process at each. Source: Google’s guidance on hiring search help and the measurement rules described in this article.

Questions about measurement

  • How do you write the questions, and do any contain the brand's name? A question that names the brand will produce the brand. In one real account a reported 13% visibility came entirely from a single brand-named question; the true unprompted figure was zero. A good provider excludes such questions from scoring. Our glossary entry on the self-referential question explains the trap.
  • What counts as a citation? Assistants typically retrieve a dozen or more pages and cite a few. Counting the retrieval pool rather than the sources marked in the answer inflates citation numbers several times over. Ask which they count.
  • Do you report per engine? Engines behave very differently. A single blended number hides where the work is.
  • How do you handle changes in your own measurement? Adding a model, switching on web search or adding questions changes the ruler, not the brand. Enabling web search once moved a site from 3% to 13% while the site did nothing. A provider should version its methodology and refuse to compare across versions.
  • What happens to a result you could not measure? A timed-out answer is not a non-mention. It should never appear as 0%.
  • How many brands are in the denominator? Being 1% of 117 brands is a different situation from 1% of five, and a share of voice figure without the brand count cannot tell them apart.

Questions about the work

  • Which of the five workstreams do you do yourselves, and which do you subcontract?
  • What will you need from us each month: approvals, facts, developer time?
  • Which changes do you propose to make to our site, and will you document each one with the reason?
  • What outreach do you do on community sites, and under whose name?
  • Can we speak to a current client in a similar category?

Google's own interview questions for an SEO are a good base here: ask for examples of previous work, what results they expect and in what timeframe, how they measure success, their experience in your industry and country, and whether they will share every change they make with the reasoning behind it. Google also suggests watching whether the provider is curious about your business: who your competitors are, what makes you different, how your customers find you. A provider that asks none of this before proposing a plan is selling a template.

Questions about access and ownership

Google advises granting a prospective SEO only read access to Search Console at first, and reminds site owners that they are ultimately responsible for the actions of any company they hire. The same caution applies to GEO providers. Grant the least access needed for each phase, keep administrative access to your CMS and community accounts with your own staff, and agree in writing that the question set, raw measurement data, content and any accounts created for the work belong to you.

Questions about pricing

Agencies price GEO as a monthly retainer, as fixed-scope projects (an audit, a measurement baseline, a content sprint) or as a mix. There is no standard rate, and published price lists are rare, so compare proposals by scope: how many questions on how many engines, how often, how many pages or outreach placements, how many hours of technical work. Be wary of pricing tied to outcomes the provider cannot control, such as payment per AI mention; it encourages measurement that finds mentions rather than measurement that is accurate.

Run a short paid trial before a long contract

If you are choosing between providers, a short paid trial on identical terms is more informative than any pitch. Give each one the same twenty or so buyer questions, approved by you, and ask for a baseline across the same engines in the same week, with raw results exported. Compare more than the headline numbers: whether the providers counted the same brands, whether they resolved spelling variants, whether they reported citations from the answer text or from everything retrieved, and whether they flagged answers that failed or timed out. Two honest measurements of the same questions should broadly agree on structure even if individual answers differ. If one provider's numbers are much better than another's on the same questions, the difference is in the method, and you now know which question to ask.

An interview scorecard

QuestionStrong answerWeak answer
Do your tracked questions mention our brand?No; brand-named questions are excluded from scoring and reported separately if at all"We track your brand terms across all the AI engines"
What counts as a citation?Only sources marked in the answer text, with the number of distinct questions shown"Every source the AI looked at"
How do you show change over time?Per engine, on a versioned methodology, trusting changes seen in consecutive runsA single score with a month-on-month arrow
Who owns the data?You do; raw results exported each run"You can see everything in our dashboard"
What results can you promise?A clean baseline, specific fixes and honest reporting; no placement promises"Top three in ChatGPT within three months"
How will you publish on our behalf?Only after named approval, logged, disclosed where required"We handle all of that so you don't have to"

GEO agency red flags

Some claims should end the conversation. They are not matters of style; each one either describes something no one can deliver, or something that will hurt the brand when it is discovered.

Guaranteed placement

"We guarantee you'll be recommended by ChatGPT" or "top three in AI Overviews in three months" is the clearest red flag. Google's guide to hiring an SEO says no one can guarantee a #1 ranking on Google, and that if someone guarantees their changes will give you first place, you should find someone else. AI answers are harder to guarantee than rankings, not easier: they change between runs of the same question, between engines and between model versions, and they depend on third-party sources nobody controls. A provider can promise work, method and honest reporting. It cannot promise an answer.

Special relationships and secret signals

Google warns against SEO providers that allege a "special relationship" with Google. The equivalent in GEO is a claim of privileged access to OpenAI, Anthropic, Perplexity or Google's AI teams, or knowledge of hidden "AI ranking factors". Ask for documentation. Where the engine has published guidance, as Google has for its AI features, compare the claim with it.

Magic files and markup

Some providers sell llms.txt files, special "AI schema" or proprietary markup as the core of the service. Structured data and a clean llms.txt can be reasonable housekeeping, but Google states that you do not need new machine-readable files, AI text files or markup to appear in its AI features, and no other major engine has documented such a file as a requirement. A proposal that leads with a file rather than with facts, access and measurement has its priorities backwards.

Volume content and fake communities

Mass-produced articles stuffed with question phrases are cheap to sell and do little; the GEO paper found keyword stuffing offered little to no improvement in generative engine responses. Worse is fabricated activity: fake reviews, sock-puppet accounts on community sites, undisclosed paid posts. These can produce a short-term mention and a long-term problem, because communities and review platforms remove them and the brand's name stays attached to the story.

AI referral traffic presented as the result

Some reports lead with sessions from AI referrers in your analytics. Those sessions are real and worth tracking, but most people who read an AI answer never click through, so referral traffic measures a small and unrepresentative slice of the effect. A provider that reports a rise in AI referral sessions as proof of visibility, without the answer-side measurement behind it, is showing you the easiest number to move rather than the one that matters.

Numbers without a denominator

A report that shows a bare score, a hypothetical "AI visibility: 64" for instance, with no question set, no engine breakdown, no date range and no brand count is not a finding. Neither is a citation count that includes every page an engine retrieved, or a jump that coincides with the provider adding a model or switching on web search. Our post on what to put in an AI visibility report covers what a defensible report contains.

Why no one can guarantee a place in an AI answerControlsYour own assetsFacts on your pagesCrawler access, renderingNaming and markupContent that answersbuyer questionsCan influenceThrough honest workCoverage on publishersDirectory and reviewprofilesCorrections of stalethird-party factsCannot controlOutside anyone’s reachRun-to-run variationModel and version updatesWhich sources are retrievedWhat competitors publishHow each engine citesA guarantee would have to cover the right-hand column.Source: this article; Google states no one can guarantee a #1 ranking in its own search results.
Figure 5. What a provider controls, can influence and cannot control. Placement in AI answers depends on the right-hand column, which is why it cannot be guaranteed. Source: this article.
Red flagWhy it is a problemWhat to ask instead
Guaranteed placement in AI answersAnswers vary by run, engine and version; no one controls them"What will you do, and how will we know whether it worked?"
Special relationship with an AI companyUnverifiable; Google explicitly warns against the equivalent claim in SEO"Which published documentation supports this?"
A special file or markup as the core deliverableGoogle says none is needed for its AI features"What facts, access fixes and sources does the plan address?"
Hundreds of question-stuffed articlesKeyword stuffing showed little to no gain in research; low-quality pages carry risk"Which buyer questions does each page answer that nothing else does?"
Community "seeding" without disclosureBreaks platform rules; discovered posts damage the brand"Under whose name, with what disclosure?"
A single blended scoreHides engines, questions and methodology changes"Show me per engine, per question, with the methodology version"

Reading what an agency reports back

Even with a good provider, the monthly report is where expectations and reality meet. Reading it critically is a skill your team should keep in-house whatever else you hand off.

Check the frame before the findings

Every report should state, before any number, which questions were run, on which engines, on what dates and under which methodology version. If any of those changed since the previous report, the comparison is not valid and should not be presented as a trend. Our guide to GEO reporting for clients sets out the frame in detail.

Look for absence, not just presence

The most useful section of a GEO report is the list of questions where you are absent and a competitor is present. That list is the brief for the next month. A report that only shows wins gives you nothing to act on, and suggests the provider is managing the relationship rather than the work.

Separate the outcomes

Named, cited and recommended are different outcomes. Your page can be cited as a source while the answer recommends a competitor. A report that merges them flatters the result and hides the gap.

Ask what the provider did not measure

Sections that could not be computed should say so. A report that quietly fills a section with plausible content is worse than a shorter report, because you cannot tell which parts to trust.

Structuring a GEO consultant or agency engagement

How the engagement is set up decides much of what you get from it. Three structures are common, and each suits a different starting point.

Audit and baseline, then decide

A fixed-scope project to measure where you stand, diagnose technical access and entity problems, and propose a prioritised plan. This is the lowest-risk way to evaluate a provider: the deliverables are concrete, the methodology is visible, and you can execute the plan in-house, with the same provider or with another. Insist that the question set and raw data are part of the deliverable.

Retainer for execution

A monthly retainer covering content, outreach and measurement operations against an agreed plan. It suits teams that have the facts and approvals in place but lack hours. Define the monthly scope in units you can verify (questions run, pages delivered, corrections secured) rather than in hours or "activity", and review the plan each quarter.

Advisory for an in-house team

A consultant who trains and reviews while your team does the work. It suits organisations that want to build the capability permanently. The measure of success is that the team needs the consultant less each quarter.

Contract clauses worth having

  • Data ownership. The question set, raw answers, citations and methodology notes belong to you and are exported after every run in a documented format.
  • Approval before publication. Nothing is published on your site, your social accounts or third-party sites in your name without written approval from a named person.
  • Disclosure. Any community or review activity carried out for you is disclosed in line with the platform's rules and the law that applies to you.
  • Methodology changes. The provider tells you in advance when it changes how it measures, and restates or separates history rather than splicing it into one trend.
  • No outcome guarantees in marketing materials. The provider does not describe your engagement publicly with placement claims you cannot verify.

Plan the exit at the start

Whatever the structure, agree at the start what you will hold when the engagement ends: the question set, the raw results of every run, the methodology notes, the content and its sources, the list of third-party contacts and corrections, and every account created for the work. If you cannot continue the measurement without the provider, you do not own your trend line.

Building GEO in-house: roles, skills and tools

If you decide to keep most of the work, the question becomes who does it. GEO rarely justifies a new department. It usually works best as a responsibility added to people who already own adjacent work, with one person accountable for the whole.

The roles you need

RoleTypical existing ownerGEO responsibilitiesNew skill to learn
GEO ownerSEO lead or content leadPlan, question set, reporting, prioritiesAnswer-side measurement and its pitfalls
TechnicalWeb developer or SEO engineerCrawler access, rendering, robots rules, datesAI crawler user agents and bot-management behaviour
FactsProduct marketingSource of truth, corrections, entity consistencyWriting facts that stand alone when lifted
ContentContent teamPages that answer buyer questionsWriting for extraction, not only for reading
CorroborationPR or communicationsCorrections and coverage on cited sourcesChoosing targets from citation data

Upskilling an SEO team

Most of the knowledge transfers. An SEO team already understands crawling, rendering, canonicalisation, structured data and content quality. The new parts are narrower: that AI crawlers are several distinct agents with different jobs, that many do not execute JavaScript, that success is measured by answers rather than positions, and that the measurement itself is easy to get wrong. A few weeks of deliberate practice, running a small question set by hand across engines and reading every answer, teaches more than any course.

Tools versus people

Measurement at any scale needs tooling: running dozens of questions across several engines every month by hand is not sustainable, and manual counting is where brand variants and retrieval pools creep in. The rest of the work needs people more than tools. When comparing tools, apply the same methodology questions you would ask an agency; a tool is an agency's measurement practice in software form, with the same pitfalls.

Our take: hire for skill, keep the truth

GEO is real work, but it is mostly an extension of disciplines that good marketing teams already practise: accurate facts, accessible pages, useful content and earned coverage. The genuinely new part is measuring answers honestly, and that is where outside expertise most often helps and most often misleads. Our view is that most companies should own their facts, their approvals, their question set and their data, and buy skill and capacity where they are short. A good agency makes that division easy; a weak one resists it, because the data and the definitions are what keep a client from leaving.

Hire an agency when you need speed, technical diagnosis or outreach volume you cannot build. Build in-house when your category is specialised and your facts change often. In either case, judge providers by their methodology and their willingness to show you where you are absent, not by a screenshot of a good answer. Guarantees of placement remain a reason to walk away.

If you decide to run the measurement yourself and keep agencies for execution, InsightWonder is one of the measurement tools built for that split; our guide to choosing a GEO tool sets out what to compare across vendors, including us.

Sources