An AI search optimization agency — worth hiring now?

Profound’s $1.8 billion valuation confirms demand is surging, but agencies still need to prove their judgment…

The short version

Profound’s $1.8 billion valuation confirms surging enterprise demand, not that AI search visibility causes sales.
Automated measurement, briefing, publication, and reporting are converting much of the traditional agency retainer into software.
Buyers should pay agencies for accountable judgment, experiment design, cross-department execution, and independent auditing rather than dashboard operation.

Profound raised $180 million less than seven months after a $96 million round, while keeping its actual revenue private. Its valuation jumped from $1 billion to $1.8 billion. Very normal startup behavior. Pass the espresso.

If I were choosing an AI search optimization agency, I would resist treating that valuation like a Michelin star. Investor pricing tells me enterprise demand is hot. It gives me zero evidence that appearing more often in ChatGPT persuaded somebody to buy a serum, book a hotel, or switch credit cards.

I initially assumed the financing would be fantastic news for agencies. Then I looked at what Profound is building and changed my mind. Its platform measures how answer engines portray a brand, finds work to do, prepares and publishes approved material, then reports on what happened. A decent chunk of the traditional agency retainer is becoming a product workflow.

The agencies left standing will need to provide judgment that software cannot quietly bundle into a dashboard.

The score depends on who is asking

An AI search optimization agency should track four things: whether an answer engine mentions a brand, which name variant it uses, how it portrays the company and what sources it cites. The useful version starts with representative prompts and ends with a business decision. A screenshot of one flattering ChatGPT response belongs in Slack beside the memes.

Bar chart comparing current figures against their baselines: Organisations with significant exposure… 86 % versus 14 %, Organisations reporting end-to-end… 14 % versus 86 %, reported order accuracy for the IBM… 80 % versus 95 %, Increase in available GPU KV-cache-token… 37 % versus 8 %.

Profound’s Answer Engine Insights shows how the machinery works. I configure prompts around questions my customers ask, and Profound sends them to selected answer engines. It captures every response and adds it to a dataset instead of treating one answer as gospel. The system maps variations of my brand name, so a weird capitalization choice from 2018 does not wreck the analysis. It records appearances, citations and the way the company is portrayed. FactCheck then compares extracted claims with my source of truth and flags citations associated with inaccuracies. Over repeated runs, I can inspect patterns instead of refreshing ChatGPT like a founder checking Stripe after launch day.

Prompt design decides what those patterns mean. Profound analyzed more than 71,000 responses from three answer engines: ChatGPT and Claude, plus Gemini. The prompts covered clothing and furniture, along with credit cards. Repeated answers for the same persona shared roughly two in five recommended brands. Different personas shared closer to one in four, a reduction of about 15 percentage points. A healthy aggregate score can therefore hide the fact that my intended buyers receive different recommendations. Profound recommends persona-level measurement, and I agree.

Retrieval can also move underneath the experiment. In Profound’s analysis of ChatGPT Shopping, feed-integrated catalog retrieval jumped from 8% to 62% in one day following the GPT-5.6 release. A merchant could wake up with dramatically different visibility before its team changed a page. Any agency reporting a smooth monthly trend needs to separate its own work from a platform change, because OpenAI does not ask permission before rearranging the furniture.

Karine Brunet described the realities of interconnected technology ecosystems:

Today’s organisations operate in highly interconnected technology ecosystems where complete independence is rarely achievable.

This exposes the limit of a visibility score. Profound’s current interface lacks a total-opportunity denominator covering responses where no brand appeared, so its Visibility Score and Mention Frequency are directional signals. Neither tells me how often my brand could have appeared across the full prompt set. They certainly cannot tell me whether a mention produced a sale.

Any agency pitching a giant score without showing the prompts and personas, explaining its entity mapping, and defining the denominator is selling numerology with a nicer font. I want to see what changed underneath the graph. Answer engines vary recommendations by audience, while a revised prompt set creates a different experiment even when the dashboard draws one continuous line.

The dashboard has learned to do the retainer

Profound says its revenue tripled over the preceding six months, according to TechCrunch. The starting amount remains private. Investors get evidence of momentum, while buyers still cannot evaluate absolute scale or profitability. Retention is private too. Multiplying a mystery number still produces a mystery number.

The more interesting signal is the product loop. Answer Engine Insights collects responses and flags weak representation, dubious claims and citations that deserve investigation. Context Manager distills knowledge-base files and internal material such as meeting transcripts or communication threads into usable company context. It also learns how an individual marketer works. AI Marketer combines that context with the answer-engine analysis and proposes projects. Its sub-agents can handle research and briefs, then prepare drafts and reviews. A marketer approves the plan before Aim carries the work through CMS publication and later reporting. When the company’s positioning changes, Context Manager updates the material guiding the next output, and another measurement cycle can show whether the answer-engine landscape moved.

That cycle is much tighter than the classic monthly retainer. I have sat through enough agency presentations to know the ritual: export data, make slides, schedule a findings call, write a brief and return later with another deck. Profound wants work to move directly from a detected problem to approved execution inside one system. The company says its latest financing will expand those capabilities and fund more applied-AI research for marketing.

Martin Merz emphasized the importance of secure and scalable sovereignty:

La résilience numérique de l’Europe repose sur une souveraineté à la fois sécurisée et scalable.

Department-store shopper holding a phone walks between clothing racks and staged furniture displays under bright retail lighting.

Profound connects answer-engine responses to company context, AI-assisted production, human approval, publication and another measurement cycle.

Calling Profound another mention tracker misses its commercial ambition. Monitoring occupies one budget line. Content production and campaign management reach much larger pools of money. Search Engine Watch argues that Profound can now compete for that spending alongside its original AI-search analytics business.

The warning for agencies is obvious. Research exports and generic content briefs are becoming buttons. Human review still matters, especially when the software produces something legally risky or magnificently cringe, but clients will stop paying premium rates for the mechanical steps between finding an issue and drafting a response.

This also changes what clients should expect from agency reporting. A monthly deck made sense when collecting evidence required manual searches, spreadsheets and several unfortunate hours of copy-pasting. Once a platform continuously captures the answers and proposes the work, the agency has to explain why an opportunity matters, what evidence would disprove its recommendation and where automation should stop. “We operated the dashboard for you” will age about as well as charging extra to email a PDF.

The valuation outran the evidence

Profound’s valuation shows that investors expect AI-mediated discovery to command serious marketing budgets. Evidence connecting Profound-driven optimization to incremental revenue remains absent. The company has disclosed neither absolute revenue nor the assumptions supporting its valuation, and outsiders cannot separate growth from monitoring and content production, let alone advertising tools or agent automation.

I will concede the strongest argument against my skepticism: marketers do not need perfect attribution before fixing obvious problems. If an answer engine repeatedly invents the wrong product price or cites a stale page, correcting the source has value. Accurate brand representation is a legitimate operational goal. Enterprise buyers can reasonably fund that work before finance calculates the last cent of return.

Sales claims require a higher standard because visibility sits several steps away from money. First, the prompt set must stay stable enough to establish a usable baseline. The optimization then has to change what the answer engine says despite normal variation between models, audience personas and retrieval behavior. A person must see the changed answer and take an action. That action enters the usual attribution swamp, where returning visitors and other marketing channels are already splashing around. Finally, the resulting revenue must exceed what probably would have happened without the intervention. Search Engine Watch says Profound’s research linked AI mentions with increased site visits, but the work lacked randomized exposure and did not measure purchases.

I am left with timing and a plausible story.

Search Engine Watch raises the same issue with Profound’s Omnilux case study. The evidence came from the vendor rather than an independent causal test, which supports a narrower claim than the triumphant slide headline probably suggests. Maybe Profound drove the outcome. The published material cannot isolate that effect from other changes happening at the same time.

We also lack production data on AI Marketer itself. Profound has not published its accuracy, error rate or average human-review burden. Nobody outside the company knows whether visibility improvements survive a model update, a retrieval overhaul or changes in the sources available to the engine. Those gaps matter because Command Center’s hours-saved metric relies on default process-time assumptions that customers can edit. An honest estimate must include human checking and corrections, plus the time spent approving work before somebody converts it into imaginary salary savings.

Buyers should treat visibility improvement as an operational result. Revenue lift remains an unproven business claim. Put them on separate lines of the procurement spreadsheet, ideally before somebody adds a rocket emoji beside both.

When I would hire the agency

I would buy the platform when my team already knows which customer questions matter, can design a stable measurement program and has people accountable for publication. I would hire an agency when I need someone to build that operating model, challenge the evidence and own the messy work across departments.

Cheap generation changes the labor equation. Agents can prepare more briefs and drafts because marginal production costs collapse. Higher volume creates more factual claims to check, followed by a queue of human approvals. Context Manager can reduce basic brand mistakes by giving agents current company knowledge. It cannot decide every legal edge case or recognize when a perfectly grammatical paragraph makes the brand sound unbearable. Review capacity becomes the constraint, especially when teams confuse “generated” with “finished.” A strong agency chooses the prompts and opportunities, then catches confident nonsense before the CMS does.

My buying questions would get specific quickly. Who chooses the prompts, and how often can they change? Can I inspect persona-level results and the sources behind inaccurate claims? Who verifies entity matching when the brand name is ambiguous? How is human editing time included in the savings estimate? What happens when visibility rises while conversions stay flat?

I would also ask the agency to explain which work it performs beyond operating Profound or a competing platform. Strategy has value. Independent measurement has value. Distribution relationships matter, and so does good editorial taste. Charging a large retainer to export a proprietary score and summarize it in Google Slides has the economic future of airport Wi-Fi.

I’d split the job rather than hand either side the whole machine. An internal team can own Profound’s context and approval loop while an agency designs experiments and audits the conclusions. It can also handle opportunities that depend on human relationships. Software provides scale; the agency earns its fee by telling the client when the dashboard is wrong.

My prediction for the end of 2027: buyers will price dashboard-only AEO retainers like commodity reporting. The premium will go to agencies willing to attach their work to an outcome they cannot quietly redefine later.

The next great AI search optimization agency will sell accountable judgment. Everyone else is renting login credentials.

Frequently asked questions

Is an AI search optimization agency worth hiring?

An AI search optimization agency is worth hiring when a company needs help designing stable experiments, selecting representative prompts, auditing evidence, and coordinating publication across departments. A platform is the better purchase when the internal team already understands customer questions, can run measurement, and owns the approval process.

What should an AI search optimization agency measure?

An AI search optimization agency should measure brand mentions, name variants, company portrayal, and cited sources across representative prompts and personas. It should also disclose entity mapping, prompt changes, and the opportunity denominator, because visibility and mention-frequency scores are directional and do not establish that a mention caused a sale.

How does Profound affect traditional agency retainers?

Profound compresses research, analysis, briefing, drafting, approval, publication, and reporting into one product workflow. That automation reduces the value of retainers built around exports and generic content briefs. Agencies must instead provide experiment design, independent measurement, editorial judgment, cross-department coordination, and accountability for outcomes.